CA2276378C - Robust digital watermarking - Google Patents
Robust digital watermarking Download PDFInfo
- Publication number
- CA2276378C CA2276378C CA002276378A CA2276378A CA2276378C CA 2276378 C CA2276378 C CA 2276378C CA 002276378 A CA002276378 A CA 002276378A CA 2276378 A CA2276378 A CA 2276378A CA 2276378 C CA2276378 C CA 2276378C
- Authority
- CA
- Canada
- Prior art keywords
- watermark
- watermarked
- data
- image
- signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/0028—Adaptive watermarking, e.g. Human Visual System [HVS]-based watermarking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/005—Robust watermarking, e.g. average attack or collusion attack resistant
- G06T1/0064—Geometric transfor invariant watermarking, e.g. affine transform invariant
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/32154—Transform domain methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/32154—Transform domain methods
- H04N1/32165—Transform domain methods using cosine transforms
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/32154—Transform domain methods
- H04N1/32187—Transform domain methods with selective or adaptive application of the additional information, e.g. in selected frequency coefficients
- H04N1/32192—Transform domain methods with selective or adaptive application of the additional information, e.g. in selected frequency coefficients according to calculated or estimated visibility of the additional information in the image
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/32267—Methods relating to embedding, encoding, decoding, detection or retrieval operations combined with processing of the image
- H04N1/32277—Compression
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/32288—Multiple embedding, e.g. cocktail embedding, or redundant embedding, e.g. repeating the additional information at a plurality of locations in the image
- H04N1/32304—Embedding different sets of additional information
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/46—Embedding additional information in the video signal during the compression process
- H04N19/467—Embedding additional information in the video signal during the compression process characterised by the embedded information being invisible, e.g. watermarking
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/80—Generation or processing of content or additional data by content creator independently of the distribution process; Content per se
- H04N21/83—Generation or processing of protective or descriptive data associated with content; Content structuring
- H04N21/835—Generation of protective data, e.g. certificates
- H04N21/8358—Generation of protective data, e.g. certificates involving watermark
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/24—Systems for the transmission of television signals using pulse code modulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2201/00—General purpose image data processing
- G06T2201/005—Image watermarking
- G06T2201/0052—Embedding of the watermark in the frequency domain
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N2201/00—Indexing scheme relating to scanning, transmission or reproduction of documents or the like, and to details thereof
- H04N2201/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N2201/3201—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N2201/3225—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title of data relating to an image, a page or a document
- H04N2201/3233—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title of data relating to an image, a page or a document of authentication information, e.g. digital signature, watermark
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N2201/00—Indexing scheme relating to scanning, transmission or reproduction of documents or the like, and to details thereof
- H04N2201/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N2201/3201—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N2201/3269—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title of machine readable codes or marks, e.g. bar codes or glyphs
- H04N2201/327—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title of machine readable codes or marks, e.g. bar codes or glyphs which are undetectable to the naked eye, e.g. embedded codes
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Editing Of Facsimile Originals (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
- Image Processing (AREA)
- Television Systems (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
A watermarking procedure that is applicable to images, audio, video and multimedia data to be watermarked divides the data to be watermarked into a set of nxn blocks, such as the 8x8 blocks of MPEG. The same watermark signal can be distributed throughout the set of blocks in a large variety of ways. This allows the insertion algorithm to be changed without affecting the decoders. The decoding procedure first sums together the DCT coefficients of N sets of 8x8 blocks to form a set of N summed 8x8 blocks and then extracts the watermark from the summed block. Since the sum of the DCT blocks is equal to the DCT of the sum of the intensity blocks, efficient decoding can occur in both the spatial and frequency domains. The symmetric nature of the decoding process allows geometric distortions to be handled in the spatial domain and other signal distortions to be handled in the frequency domain. Moreover, insertion of a watermark signal into image data and the subsequent extraction of the watermark from watermarked image data which has been subject to distortion between the times of insertion and extraction involves the insertion of multiple watermarks designed to survive predefined distortions of the image data, such as panscan or letterbox mode transformations. Alternatively, a registration pattern in the image data, after the image data containing the registration pattern is subject to an unknown distortion, is used to compensate for distortion of the watermarked image data.
Description
r ROBUST DIGITAL WATERMARKING
Background of the Invention 1. Field of the Invention:
The present invention relates to digital watermarking of data including image, video and multimedia data. Specifically, the invention relates to insertion and detection or extraction of embedded signals for purposes of watermarking, in which the insertion and detection procedures are applied to sums of subregions of the data. When these subregions correspond to the 8x8 pixel blocks used for MPEG (moving picture experts group) and JPEG (joint photograph experts group) compression and decompression, the watermarking procedure can be tightly coupled with these compression algorithms to achieve very significant savings in computation. The invention also relates to the insertion and detection of embedded signals for the purposes of watermarking, in which the watermarked data might have undergone distortion between the times of insertion and detection of the watermark.
Background of the Invention 1. Field of the Invention:
The present invention relates to digital watermarking of data including image, video and multimedia data. Specifically, the invention relates to insertion and detection or extraction of embedded signals for purposes of watermarking, in which the insertion and detection procedures are applied to sums of subregions of the data. When these subregions correspond to the 8x8 pixel blocks used for MPEG (moving picture experts group) and JPEG (joint photograph experts group) compression and decompression, the watermarking procedure can be tightly coupled with these compression algorithms to achieve very significant savings in computation. The invention also relates to the insertion and detection of embedded signals for the purposes of watermarking, in which the watermarked data might have undergone distortion between the times of insertion and detection of the watermark.
2. Description of the Prior Art:
The proliferation of digitized media such as image, video and multimedia is creating a need for a security system that facilitates the identification of the source of the material. , Content providers, i.e. owners of works in digital data form, have a need to embed signals into '?4570-73 video/image/multimedia data, which can subsequently be detected by software, and/or hardware devices for purposes of authentication of copyright ownership, and copy control and management.
For example, a coded signal might be inserted in data to indicate that the data should not be copied. The embedded signal should preserve the image fidelity (image quality), be robust to common signal transformations and resistant to tampering. In addition, consideration must be given to the data rate that can be provided by the system, though current requirements are relatively low --a few bits per frame.
In U.S. Patent 5,930,369, July 27, 1999, entitled "Secure Spread Spectrum Watermarking for Multimedia Data", there was proposed a spread spectrum watermarking method which embedded a watermark signal into perceptually significant regions of an image for the purposes of identifying the content owner and/or possessor. A strength of this .approach is that the watermark is very difficult to remove. In fact, this method only allows the watermark to be read if the original image or data is available for comparison. This is because the original spectrum of the watermark is shaped to that of the image through a non-linear multiplicative procedure, and this spectral shaping must be removed prior to detection by matched filtering. In
The proliferation of digitized media such as image, video and multimedia is creating a need for a security system that facilitates the identification of the source of the material. , Content providers, i.e. owners of works in digital data form, have a need to embed signals into '?4570-73 video/image/multimedia data, which can subsequently be detected by software, and/or hardware devices for purposes of authentication of copyright ownership, and copy control and management.
For example, a coded signal might be inserted in data to indicate that the data should not be copied. The embedded signal should preserve the image fidelity (image quality), be robust to common signal transformations and resistant to tampering. In addition, consideration must be given to the data rate that can be provided by the system, though current requirements are relatively low --a few bits per frame.
In U.S. Patent 5,930,369, July 27, 1999, entitled "Secure Spread Spectrum Watermarking for Multimedia Data", there was proposed a spread spectrum watermarking method which embedded a watermark signal into perceptually significant regions of an image for the purposes of identifying the content owner and/or possessor. A strength of this .approach is that the watermark is very difficult to remove. In fact, this method only allows the watermark to be read if the original image or data is available for comparison. This is because the original spectrum of the watermark is shaped to that of the image through a non-linear multiplicative procedure, and this spectral shaping must be removed prior to detection by matched filtering. In
3 addition, the watermark is usually inserted into the N
largest spectral coefficients, the ranking of which is not preserved after watermarking. This method does not allow software and hardware devices to directly read embedded signals without access to the original unwatermarked material.
In an article by I.J. Cox, J. Kilian, 'T. Leic~hton and T.
Shamoon, "A Secure, Robust Watermark for Multimedia", Information Hiding, LNCS 1174, Springer-Verlag, pp. 185-206 (1996) spread spectrum watermarking is described which embeds a pseudo-random noise sequence into the digital data for watermarking purposes.
The above prior art watermark extraction methodology requires the original image spectrum be subtracted from the watermark image spectrum., This restricts the use of the method when there is no original image or original image spectrum available to the decoder.
One application where this presents a significant difficulty is for third party device providers desiring to read embedded information for operation or denying operation of such a device.
In U.S. Patent No. 5,319,735 by R. D. Preuss et al.
entitled "Embedded Signaling" digital inform<~tion is encoded to produce a sequence of code symbols. The sequence of code symbols is embedded in an audio signal by generating a corresponding sequence of spread spectrum code signals representing the sequence of code symbols.
largest spectral coefficients, the ranking of which is not preserved after watermarking. This method does not allow software and hardware devices to directly read embedded signals without access to the original unwatermarked material.
In an article by I.J. Cox, J. Kilian, 'T. Leic~hton and T.
Shamoon, "A Secure, Robust Watermark for Multimedia", Information Hiding, LNCS 1174, Springer-Verlag, pp. 185-206 (1996) spread spectrum watermarking is described which embeds a pseudo-random noise sequence into the digital data for watermarking purposes.
The above prior art watermark extraction methodology requires the original image spectrum be subtracted from the watermark image spectrum., This restricts the use of the method when there is no original image or original image spectrum available to the decoder.
One application where this presents a significant difficulty is for third party device providers desiring to read embedded information for operation or denying operation of such a device.
In U.S. Patent No. 5,319,735 by R. D. Preuss et al.
entitled "Embedded Signaling" digital inform<~tion is encoded to produce a sequence of code symbols. The sequence of code symbols is embedded in an audio signal by generating a corresponding sequence of spread spectrum code signals representing the sequence of code symbols.
4 The frequency components of the code signal being essentially confined to a preselected signaling band lying within the bandwidth of the audio signal and successive segments of the code signal corresponds to successive code symbols in the sequence. The audio signal is continuously frequency analyzed over a frequency band encompassing the signaling band and the code signal is dynamically filtered as a function of the analysis to provide a modified code signal with frequency component levels which are, at each time instant, essentially a preselected proportion of the levels of the audio signal frequency components in corresponding frequency ranges.
The modified code signal and the audio signal are combined to provide a composite audio signal in. which the digital information is embedded. This component audio signal is then recorded on a recording medium or is otherwise subjected to a transmission channel. Two key elements of this process are the spectral shaping and spectral equalization that occur at the insertion and extraction stages, respectively, thereby allowing the embedded signal to be extracted without access to the unwatermarked original data.
In U.S. Patent No. 5,848,155 by Cox entitled "A
Spread Spectrum Watermark for Embedded Signaling~'~
there is described a method for extracting a watermark of embedded data from watermarked images or video without using an original or unwatermarked version of the data.
This method of watermarking an image or image data for embedded signaling requires that the DCT (discrete cosine transform) and its inverse (IDCT) of the entire
The modified code signal and the audio signal are combined to provide a composite audio signal in. which the digital information is embedded. This component audio signal is then recorded on a recording medium or is otherwise subjected to a transmission channel. Two key elements of this process are the spectral shaping and spectral equalization that occur at the insertion and extraction stages, respectively, thereby allowing the embedded signal to be extracted without access to the unwatermarked original data.
In U.S. Patent No. 5,848,155 by Cox entitled "A
Spread Spectrum Watermark for Embedded Signaling~'~
there is described a method for extracting a watermark of embedded data from watermarked images or video without using an original or unwatermarked version of the data.
This method of watermarking an image or image data for embedded signaling requires that the DCT (discrete cosine transform) and its inverse (IDCT) of the entire
5 image be computed. There are fast algorithms for computing the DCT in N log N time, where N is the number of pixels in the image. However, for N=512x512, the computational requirement is still high, particularly if the encoding and extracting processes must occur at video rates, i.e. 30 frames per second. This method requires approximately 30 times the computation needed for MPEG-II
decompression.
One possible way to achieve real-time video watermarking is to only watermark every N-th frame.
However, content owners wish to protect each and every video frame. Moreover, if it is known which frames contain embedded signals, it is simple to remove those frames with no noticeable degradation in the video signal.
An alternative option is to insert the watermark into nxn blocks of the image (subimages) where n«N. If the block size is chosen to be 8x8, i.e. the same size as that used for MPEG image compression, then it is possible to tightly couple the watermark insertion and extraction procedures to those of the MPEG compression and decompression algorithms. Considerable computational saving can then be achieved since the most expensive computations relate to the calculation of the DCT and its
decompression.
One possible way to achieve real-time video watermarking is to only watermark every N-th frame.
However, content owners wish to protect each and every video frame. Moreover, if it is known which frames contain embedded signals, it is simple to remove those frames with no noticeable degradation in the video signal.
An alternative option is to insert the watermark into nxn blocks of the image (subimages) where n«N. If the block size is chosen to be 8x8, i.e. the same size as that used for MPEG image compression, then it is possible to tightly couple the watermark insertion and extraction procedures to those of the MPEG compression and decompression algorithms. Considerable computational saving can then be achieved since the most expensive computations relate to the calculation of the DCT and its
6 inverse and these steps are already computed as part of the compression and decompression algorithm. The incremental cost of watermarking is then very small, typically less than five percent of the computational requirements associated with MPEG.
U.S. Patent 6,069,914 issued May 30, 2000, entitled "Watermarking of Image Data Using MPEG/JPEG Coefficients", advances this work by using MPEG/JPEG coefficients to encode the image data.
U.S. Patent 5,91~i,027 issued June 22, :L999, entitled "Digital Watermarking", describes storing watermark information into subimages and extracting watermark information from subimages.
A review of watermarking is found in an article by Cox et al., entitled "1?, review of watermarking and the importance of perceptual modeling" in Proc. of EI'97, vol.
30-16, February 9-14, 1997.
There have been several proposals to watermark MPEG
video or JPEG compressed still images. In a1.1 cases, each 8x8 DCT block is modified to contain the watE~rmark or a portion thereof. Consequently, decoding of the watermark requires that each 8x8 block be individually analyzed to extract the watermark signal contained therein. The individual extracted signals may then be combined to form a composite watermark, which is then compared with known ' ' 7 watermarks. Because each block must be analyzed , individually, an uncompressed image must be converted back to the block-based DCT representation, which is computationally expensive. Thus, while the decoder may be computationally efficient in the DCT domain, extracting a watermark from the spatial domain is much more expensive.
To allow for computationally efficient detection of the watermark in both the spatial and DCT domains, a watermark may be inserted in the sum of all the 8x8 blocks in the DCT domain, or the sum of a subset of all the 8x8 blocks in the DCT domain. A major advantage of this approach is that if the image is only available in the spatial domain, then the summation can also be performed in the spatial domain to compute a small set of summed 8x8 blocks and only those blocks must then be transformed into the DCT domain. This is because the sum of the DCT blocks is equal to the DCT of the sum of the intensities. Thus, the computational cost of decoding in the DCT and spatial domains is approximately the same.
A second advantage of watermarking the sum of the DCT blocks is that there are an unlimited number of equivalent methods to apportion the watermark throughout the image. For example, if the watermark requires a change of 0i to the i-th coefficient of the summed DCT
block, then, if there are M blocks in the image, ~i/M can be added to each individual block, or block 1 can have ~i added to it and the remaining M-1 blocks left unaltered, ignoring for the moment issues of image fidelity. Because of this one to many mapping, it is possible to alter the insertion algorithm without changing the decoder. This is a very important characteristic, since in some watermarking applications, there may be many hardware decoders that are deployed, such that changing the decoder is impractical. However, improvements to the insertion algorithm can still result in improved detection using the approach described herein.
A third advantage of watermarking the sum of the DCT blocks is that watermark signals extracted from these sums have small variances, compared with the amount that they may be changed without causing fidelity problems.
This means that, in many cases, it is possible to change an image so that the summed DCT blocks perfectly match the required watermark signal, even though the resulting image appears identical to the original.
Finally, it is well known that some problems, such as modeling the human visual system, are best performed in the frequency domain, where other problems such as geometric transformations are more conveniently dealt with in the spatial domain. Since the computational cost of decoding the watermark is now symmetric, it is possible to switch from spatial to frequency domains at will in order to correct for various signal transformations that may corrupt the watermark.
Summary of the Invention In accordance with the present invention, there is provided a method of inserting a watermark signal into data to be watermarked comprising the steps of: applying a watermark extraction to unwatermarked data for generating extracted values; computing a difference between a watermark to be inserted and the ~~xtracted values; and adding the computed difference throughout the unwatermarked data.
In accordance with the present invention, there is also provided a method of detecting a watermark from watermarked data comprising the steps of: receiving 'watermarked data where the watermarked data comprises nxn :blocks of watermarked data; summing the nxn blocks of watermarked data to form at least one nxn blo~~k of summed watermarked data; cyclically shifting the blo~~k in a spatial domain to register the block; and extracting 'the watermark from the at least one n:~n block.
In accordance with the present invention, there is <also provided a method of detecting a watermark from watermarked data comprising the steps of: receiving watermarked data where t=he watermarked data comprises nxn blocks of watermarked data; summing the nxn blocks of watermarked data to forrn at least one nxn block of summed watermarked data; and extracting the watermarlt from the at :Least one nxn block by summing predetermined watermarked data from each nxn block to yield each elements of the extracted watermark.
In accordance with the present invention, there is also provided a method of detecting a waterma:rk from watermarked data comprising the steps of: receiving 9a watermarked data where the watermarked data comprises nxn blocks of watermarked data; summing the nxn blocks of watermarked data to form M nxn blocks of summed watermarked data, where M is number of functions mapping frequency indices onto indices of elements of watermarks; and extracting the watermark from the at least one nxn block.
In accordance with the present invention, there is also provided a method of inserting a watermark signal into video images comprising the steps of: receiving input video images; performing discrete cosine transformation (DCT) of said input video images to obtain input video image DCT
'values; adding watermark signals to said input video image DCT values to obtain watermarked DCT values; performing an inverse DCT on the watermarked DCT values for generating watermarked images; subtracting predicted ima!~es stored in memory from said watermarked images for generating residual images; performing discrete cosine transformation of said residual images to obtain residual image DCT values and quantizing the residual image DCT values to d~°rive residual image quantized DCT values; inverse quantizing the residual :image quantized DCT values to obtain residual image de-quantized DCT values; performing discrete cosine transformation of the predicted images to obtain predicted :image DCT values; summing the residual image de-quantized DCT values and the pred_~cted image DCT values for generating decoded image DCT coefficients; calculating correction signals by applying a watermark extraction to the decoded :image DCT coefficients and comparing resultani~ extracted values with watermark target values; adding the correction signals to the residual image quantized DCT values for obtaining an output signal; inverse quantizing the output signal to obtain inverse quantized output data, performing inverse DCT of the inverse quantized output signal and 9b summing the resultant signal with the predicted images for generating a summed signal; generating predicted images to be stored in said memory by using the summed signal; and variable length encoding the output signal for providing a watermarked MPEG video signal of the input video images.
In accordance with the present invention, there is also provided a method of inserting a watermark signal into video images comprising the steps of: receiving input video images; performing discrete cosine transformation (DCT) of said input video images to obtain input video image DCT
values; adding watermark signals to said input video image DCT values to obtain watermarked DCT values; performing discrete cosine transformation on predicted images stored in memory to obtain predicted image DCT values; subtracting the predicted image DCT values from the watermarked DCT values to obtain residual image DCT values; quantizing the residual image DCT values to obtain residual image quantized DCT
values; inverse quantizing the residual image quantized DCT
values to obtain residual image de-quantized DCT values and summing the residual image de-quantized DCT values with the predicted image DCT values, for generating decoded image DCT
coefficients; calculating correction signals by applying a watermark extraction to the decoded image DCT coefficients and comparing resultant extracted values with watermark target values; summing the correction signals with the residual image quantized DCT values to obtain an output signal; inverse quantizi.ng the output signal to obtain inverse quantized output. signal, performing inverse DCT of t:he inverse quantized output signal and summing the resultant signal with tree predicted images to provide locally-decoded images; generating predicted images to be stored in said memory using the locally-decoded images; and 9c valuable length encoding the output signal to provide a watermarked MPEG video signal of the input video images.
In accordance with the present invention, there is also provided a method ~f inserting an extractable watermark signal into image data so that the watermark signal can be extracted after the watermarked data is subject to distortion comprising the steps of: providing image data to be watermarked; copying the image data; applying a predefined distortion to the copy of image data; inserting a watermark signal into the distorted copy of image data;
subtracting the copy of distored image data from the watermarked copy of distorted image data to generate a watermark pattern; performing inverse distortion on the watermark pattern to generate the watermarked pattern for nondistorted image data; and combining said image data to be watermarked and said watermark pattern for nondistorted image data to obtain watermarked data.
In accordance with the present invention, there is also provided a method of inserting an extractable watermark signal into image data so that the watermark ;signal can be extracted after the watermarked data is subject to distortion comprising the steps of: receiving image data in form of nxn blocks; and inserting an nxn registration pattern into the image data.
In accordance with the present invention, there is also provided a method of detecting a watermark from watermarked image data that has been subjected to a two-dimensional translation distortion comprising the steps of:
partitioning received distorted watermarked image data into a set of nxn blocks; summing together watermarked image data in a predetermined number of nxn blocks; if summed nxn blocks are not in spatial domain, transforming summed nxn 9d blocks into spatial domain; determining two-dimensional translation distortion of the summed nxn. blocks; shifting the summed nxn blocks according to the determined distortion; transforming the shifted nxn blocks into transform domain; extracting a possible watermark signal from the transformed nxn blocks; correlating the possible watermark with a known watermark; repeating said correlating for cyclic shifts of the possible watermark; and determining the presence of a watermark based on said correlating.
In accordance with the present invention, there is also provided a method of detecting a watermark from watermarked image data that has been subjected to a two-dimensional translation distortion comprising the steps of:
(a) partitioning received distorted watermarked image data :in nxn blocks; (b) summ_Lng together watermarked image data in a predetermined number of nxn blocks; (c) if summed nxn blocks are not in spatial domain, transforming summed nxn blocks into spatial domain; (d) shifting the summed nxn blocks by one of n2 possible shifts; (e) tran~;forming the shifted nxn blocks into transform domain; (f) extracting a possible watermark signal from the transformed nxn blocks;
(g) correlating the possible watermark with a known watermark; (h) repeating said correlating for cyclic shifts of the possible watermax-k; (i) repeating steps (d) - (h) for each of the n2 possible shifts; and (j) determining the presence of a watermark based on the results of step (i).
~e The present invention concerns a novel insertion method which employs a specific model of the human visual system which provides much better control over image fidelity. Tests have shown that it is possible to obtain large signals (more than 15 standard deviations fzom 0 correlation) with images that are indistinguishable from their respective original images.
The method handles robustness against various types of attacks in ways that are easy to relate to the specific type of attack.
The method is adaptable so that the model of the human visual system and the techniques used for handling attacks can be changed later without having to change the detector. The result is that it is possible to continue improving watermarking, particularly DVD (digital video disk) watermarking, even after many detectors have been installed. This is analogous to the situation with MPEG
video for which encoder technology can be improved without having to change existing decoders.
Use of the present insertion method allows a simple detection algorithm in either MPEG or decompressed domains.
The invention also concerns a novel detection method which is easy to implement, easy to analyze and has a low computational cost, whether the incoming video is MPEG compressed or uncompressed.
The present invention also concerns a novel insertion method that hides multiple patterns in the data.
These patterns fall into two categories: 1) registration 5 patterns used during detection to compensate for translational shifts, and 2) watermark patterns that encode the information content of the watermark.
A principal object of the present invention is the provision of a digital watermark insertion method which 10 allows detection of watermarks after the watermarked data is subjected to predefined scale changes, without modification to the watermark detector.
Another object of the invention is the provision of a watermark detection method that is computationally inexpensive in either the MPEG or decompressed domains.
A still other object of the invention is the provision of a digital watermarking method that withstands attacks without having to change a detector.
Further and still other objects of the invention will become more clearly apparent when the following description is read in conjunction with the accompanying drawings.
Brief Description of the Drawings Figure 1 is a general data flow diagram of a method of inserting a watermark into media data;
Figure 2 is a schematic diagram of an MPEG-2 encoder;
' '. 11 Figure 3 is a schematic diagram of a modified MPEG-2 encoder for reducing degradation of a watermark in watermarked data;
Figure 4 is a schematic diagram of an alternative modified MPEG-2 encoder for reducing degradation of a watermark in watermarked data;
Figure 5 is a flow chart of the process performed in the watermark correction device in Figures 3 and 4;
Figure 6 is a flow chart of the process performed in step 506 of Figure 5;
Figure 7 is a flow chart of an alternative process performed in step 506 of Figure 5;
Figure 8 is a flow diagram of a method of extracting a watermark from MPEG media data;
Figure 9 is a flow diagram of a method of extracting a watermark from uncompressed media;
Figure 10 is a flow diagram of a method of detecting a watermark from MPEG data, with registration;
and Figure 11 is a flow diagram of a method of detecting a watermark from uncompressed image data, with registration.
Detailed Description of the Preferred Embodiment As used in the following description the terms image and image data will be understood to be equally applicable to video, image and multimedia data. The term "watermark" will be understood to include embedded data, symbols, images, instructions or any other identifying information.
In order to better understand the present invention, first a review of the basic watermarking method will be presented followed by additional descriptions of the improvements comprising the present invention.
First, we define some notations. Let a watermark to be embedded into an image be an N dimensional vector, denoted by W[1,...,N]. In the following text, the notation W[1,...,N] is used in the same manner as W[k]
(k=1,...,N). Let V[1,...,N] denote a vector value extracted from an image, where the element V[k]
corresponds to W[k]. Specifically, the value V[k] is a weighted sum of DCT coefficients given by V[k]=D k[1]F k[1]+D k[2]F-k[2]+...+D k[n k]F k[n k], where D k[i] (i=1,...,n_k) indicate members of the set of DCT coefficients used for calculating V[k], n k indicates the number of members, and F k[i] (i=1,...,n k) are weighting coefficients related to a filter processing.
The concept of F k is that the DCT coefficients are weighted according to how much noise might be expected in each coefficient. To calculate V[1,...,N], nxn DCT
coefficients are first calculated over a whole image.
Then the coefficients are classified into N sets, each of which is related to each element of V[1,...,N]. The rule of classifying DCT coefficients is predetermined, and the same rule is used in both inserting and detecting a watermark.
Before inserting a watermark into an image, a detection algorithm is applied to the image to find a watermark that is already present in the image. If the image does not contain a watermark, the extracted values, V[1,...,N] will be normally distributed random numbers °
that do not correlate any watermark W[1,...,N]. A
watermark W[1,...,N] is inserted into an image by changing each of D k[1,...,n k] (k-i,...,N) slightly in order to make the extracted value V[1,..., N] highly correlate the watermark W[1,...,N]. Let the target value of V[1,...,N] be denoted by w[1,...,N], that is, the values V[1,...,N] are changed to.w[1,...,N] by inserting the watermark W[1,...,N]. The target values w[1,...,N]
have high correlation with the watermark W[1,...,N] and they are determined as will be described below.
After the target values ~[1,...,N] are determined, the difference w[k]-V[k] is distributed among the DCT
coefficients D k[1,...,n k]. Then a watermark is inserted by adding the allocated difference value to the corresponding DCT coefficients D k[1,...,n k]. This change of DCT coefficients must be done in such a manner so as not to change the appearance of the image.
In distributing the difference w[k]-V[k], a characteristic of the human visual process is taken into ' '. 14 account. The amount of change that does not cause a visible change in the image is different for each DCT
coefficient D k[i]. This amount depends on the human visual process which can be approximately simulated with a computational model. The amount of change is referred to as "slack". The slack is calculated for each DCT
coefficient and it is used in distributing the difference w[k]-V[k] among the DCT coefficients. Next we describe how to calculate the values of the slack using a model of the human visual process. .
The preferred computational model of human visual sensitivity that is used in the present invention is found in an article by Andrew B. Watson, entitled "DCT
Quantization Matrices Usually Optimized for Indirect Images" in SPIE, vol. 1913 (1993), pp. 2U2-216. This model was applied to watermarking in an article by Christine I. Podilchuk and Wenjun Zeng entitled "Digital Image Watermarking Using Visual Models", Proc. of EI'97, vol. 3016, February 9-14, 1997. The current invention differs from that of Podilchuk and Zeng in (i) not requiring the original unwatermarked image at the decoder and (ii) not extracting the watermark from the individual 8x8 blocks, but from the sum of a set of 8x8 blocks.
Other computational models are also usable.
For each element of the image's block DCT, d[i,j], this model computes a value called the element's "slack", S[i,j], which indicates how much a particular d[i,j]
15 , value may be altered before such an alteration becomes visible. The value is computed in three steps. The first step models the contrast masking phenomenon of the human visual system and models the visual sensitivity at different frequencies and handles the difference between visual sensitivity to changes in different frequencies.
The second step models the luminance masking phenomenon of the human visual system and handles the fact that the visual system is more sensitive to changes in dark regions than to changes in bright regions. The third step handles the fact that the sensitivity to changes depends in part on the percentage that the frequency is changing (i.e. a DCT term with a small value in it may only change a little, while one with a larger value may change more).
The perceptual model makes use of a matrix of values that indicate the relative sensitivity of the human visual system to the different terms of a spatial 8x8 DCT. The formulae for computing this matrix are available in an article by Albert J. Ahumada Jr. and Heidi A. Peterson entitled "Luminance-Model-Based DCT
guantization for Color Image Compression", in SPIE, vol.
1666, (1992) pp. 365-374.
After computing the slacks for all the 8x8 DCT's in the image, a slack can be assigned to each D k[1,...,n k].
Call these slacks S k[1,...,n k]. It is now possible to distribute the changes in the v's over all the D k's with minimal visual impact. This is done according to the following formula:
v,_l~[1]-D-h[1]+(~[h1]-Y[k])*S l~[i] .
~S-l~[j]F-k[j]
where D' k[1,...,n k] are the modified 8x8 DCT
coefficients, and co[k] and V[k] are the k-th elements of cu and V, respectively. The effect of this formula is to distribute the desired change in a given element of the watermark vector (o~[k] - v[k]) over all the DCT
coefficients that are summed at calculation of V[k] to produce that element, proportionately according to those DCT coefficients' slacks. To illustrate, consider two simple examples: 1) If all the slacks are 0 except for slack S k[m], then the sum of all the values S k[j]F k[j]
is equal to S k[m]F k[m], and only D k[m] is changed. It is changed by the full value of (c~[k]-V[k]). 2) If all the slacks are equal and all the coefficients F k[i]=1, then each D k[i] is changed by the same amount.
After making these changes, convert all the 8x8 DCT's back into the spatial domain, and the result is a watermarked image. It is easy to show that the sum of all the D-k[i]F k[i]'s for a given k will equal ~[k]. The process of making this is referred to as "inserting Omega into the image". The watermark extracted from the resulting image, if the image has not been attacked, will be exactly w, not w plus noise.
There are two important issues remaining to be discussed. First, how to decide on w, and, second, how to make the watermark robust.
Previously, the equivalent of co was computed as:
~ = V + a*W
where a is a small constant, and W is a zero-mean watermark signal. It is possible to use the same formula here, but it is too limiting to result in the strongest possible watermark using the present invention. In practicing the invention, it is often possible to insert an w that has perfect correlation with the watermark, W, without causing any visible change in the image. The following formula is used:
w = mean(V) + (3*(v - mean(v)) + a*W
This result is a weighted sum of the watermark signal and the original, noise (image) signal. If (3 is set to 0, the result is an cu that perfectly correlates with W.
The signal to noise ratio (SNR) for an unattacked image will be:
SNR = a*std(W)/(3*std(V) where std(X) is the standard deviation of X.
There are many ways to choose a and (3 based on optimizations to maximize different criteria such as fidelity or robustness.
'. 18 At this point there is a complete method of inserting watermarks. The method contains explicit modeling of human vision, but it does not contain any explicit method of making the watermark robust. In fact, the method as described so far will try to put as much of the watermark as possible into the high frequencies since these frequencies have the largest slack, but this is a poor thing to do from the point of view of robustness.
To make the watermark robust against a given set of , attacks or signal degradations, it is first necessary to consider how those attacks affect the various terms of the 8x8 DCT's in the image. Then, terms that are affected by attacks or signal degradations in similar ways are grouped together, and watermarked as if the group of terms were a separate image.
The following is a simple example. Suppose there is only concern about two possible attacks: cropping 24 columns of pixels from the left side of the image, or cropping 24 columns of pixels from the right side of the image. This results in three groups of DCT terms: those that come from the 3 left-most columns of 8x8 DCT's blocks, those that come from the 3 right-most columns, and those that come from the rest of the image. All the terms in each of these groups either survives or is destroyed by any given attack together. If each group is watermarked as though it were a separate image, then the watermark from at least one group will generally survive ' '. 19 attack (assuming that the 24-column cropping attacks are the only attacks possible), and the watermark that is extracted will consist of the correct watermark, from that group, plus some noisy watermar)cs from groups that were damaged by the attack.
A more interesting example is low-pass and high-pass filtering attacks. It is possible to group all the low frequencies together into one group, and all the high frequencies into one or more other groups. If the predetermined rule for classifying DCT coefficients into the N sets is designed in such a way that each set has coefficients of many different frequencies, then the complete watermarlc can be inserted into each group. Then, if the high frequencies are removed, the watermark will still be detectable in the low frequencies, and vice-versa.
The more groups the terms are divided into, the more robust will be the watermark. There is a cost because it will become increasingly more difficult to distribute the changes without causing visible distortion.
It is important to note that the best balance can be achieved after the detectors are in wide use. It is possible to modify the insertion algorithm to make watermarks robust against a wide variety of attacks without having to change detection at all.
Presently only three groups are used. One group collects together most of the low frequencies. Each of '. 20 the other two groups represents one higher frequency.
In the preferred method, a watermark is not placed in any of the higher frequencies. The reason is not that they are susceptible to attack (that is handled by the design of the filter used in detection). Rather, it is because watermarking the higher frequencies causes MPEG
compression rates to go down substantially. An alternative solution might be to add other groups that contain higher frequencies.
In Figure 1, there is shown a flow diagram of the watermark insertion method. The digital image is divided into a collection of nxn blocks, preferably 8x8 blocks, in step 10. The discrete cosine transform (DCT) of each block is computed in a known manner in step 12. The DCT's are separated into groups that respond to different attacks in the same manner in step 14. A first group G is selected in step 16.
Next, extract a watermark V, using only the DCT
terms in the group G in step 18. Determine a new signal (target value) w selected such that cu is similar to V but is highly correlated with watermark W in step 20.
Add fractions of w-V to the DCT terms in G
according to perceptual slack in step 22. Decide whether group G is the last group in step 24. If not, select next group G in step 26, and extract watermark V using only terms of next group G in step 18 and continue procedures until the last group G is found in step 24. Next, compute ' 21 the inverse DCTs of the blocks in step 28 resulting in a watermarked image.
Some alternative steps in the insertion method are possible. For example, the distribution of the difference between cu and V over DCT terms can be done stochastically to help deter tampering and reduce susceptibility to tampering.
Also, the groups of DCT terms for robustness purposes could be performed dynamically. For example, the insertion program could simulate various attacks on the image and determine the effect on the values of the DCT
term. Then, the program would cause appropriate allocation of the terms into the groups similarly affected.
The inserter can be designed with a user-interface that allows the user to set two parameters: (1) the maximum perceptual difference (J) between the original image and the watermarked image and (2) the maximum allowable probability of missed detection after any of , the predefined set of attacks. The algorithm would then insert watermarks into a large number of images automatically, according to the allowable perceptual change J and checking each one against simulations of the attacks. If an image fails to meet the specified robustness constraint (maximum allowable probability of missed detection), then the user would be notified so that a manual decision can be made to compensate or trade-off image fidelity for robustness.
In addition, the distribution of the difference between w and V over DCT terms can be modified to explicitly compensate for MPEG quantization. Using the above watermark insertion method there may result a degraded watermark in the watermarked data. In order to enhance the watermark in the watermarked data after MPEG
compression several techniques are possible.
Figure 2 is a schematic diagram of a typical MPEG-2 encoder. Figure 2 depicts elements which are indispensable to execute an MPEG-2 encoding of P pictures, or to perform a combined interframe prediction and DCT
coding. Input images are provided as one input to subtractor 30. The other input to subtractor 30 is predicted image generated in frame memory 32. The predicted images are subtracted from the input images at subtractor 30. A discrete cosine transform (DCT) is performed at DCT calculator 34 on the output signal from subtractor 30. The DCT coefficients are quantized in quantizer 36. The outputs of the quantizer 36 are sent to a variable length encoder 38 where Huffman encoding is performed. The quantized DCT coefficients outputted from the quantizer 36 are also sent to an inverse quantizer 40 where they are de-quantized. Inverse DCT of the de-quantized DCT coefficients is performed in the inverse DCT calculator 42. The results are added at adder 44 to the predicted image outputted from the frame memory 32, and then an image which is expected to be the same as that acquired in a decoder is reconstructed. The reconstructed image is called "a locally decoded image."
This locally decoded image is stored in the frame memory 32 to produce the predicted images.
Figure 3 is a schematic diagram of a modified MPEG-2 encoder for reducing degradation of a watermark in watermarked data. Before MPEG-2 encoding, DCT is performed on a input image at the DCT calculator 50 and watermark signals are added to the DCT coefficients at adder 52. The output DCT coefficients including watermark information is subject to inverse DCT in the inverse DCT
calculator 54. The output of inverse DCT calculator 54 are images with a watermark. These watermarked images are sent to MPEG-2 encoder and MPEG-2 encoding is performed as described above. In addition, in this embodiment, watermark information is modified in order to be suited to MPEG-2 compression. DCT coefficients for the predicted images are calculated in DCT calculator 56. The quantization values outputted from quantizer 36 are de-quantized in inverse quantizer 58. The results of inverse quantization are added at adder 60 to the DCT
coefficients outputted from DCT calculator 56. The results of addition correspond to the DCT coefficients for the decoded images which are expected to be generated in a decoder. These DCT coefficients are inputted into a watermark correction device 62. The watermark correction '. 24 device 62 outputs watermark correction signals. At adder 64, the watermark correction signals from device 62 are added to the quantization values from quantizer 36. The output of adder 64 is used as the inputs to variable length encoder 38 and inverse quantizer 40.
Next, we describe the process performed in the watermark correction device 62. Let us introduce several new notations to explain the process. Let Dq k[i] be the quantization value corresponding to D k[i], that is, the quantization value of the i-th member of the k-th set for calculating the value V. Let Q k[i] be the quantization step size used in obtaining Dq k[i]. Let Dr k[i] be the output value of adder 60 that is obtained by adding the inverse quantization value of Dq k[i] calculated in inverse quantizer 58 to the corresponding DCT coefficient outputted from DCT calculator 56. Let Vr[1,...,N] be the value extracted from the output values of adder 60, Dr k[i], in the same manner that the value V[1,...,N] is calculated in inserting a watermark. We assume that the target value w[1...,N] is also available in the watermark correction device 62.
Figure 5 shows the flow chart describing the process performed in the watermark correction device 62.
First the index k of the watermark element is set to 1 at step 500. Next the value Vr[k] is calculated at step 502 by Vr[k] - F k[1]Dr k[1] +...+ F k[n_k]Dr k[n k], where the weighting coefficients F k[1,...,n k] are the same value as those used in calculating V[k]. Then the absolute value of the difference between the value Vr[k]
and the target value co[k], and the sign of the difference 5 are computed in step 504 by the following equations:
Dif = ~Vr[k]-w[k]~
s = Sign(Vr[k]-cu[k] ), where 1 ( x >_ 0) S1~'11 (X ) -1 (x<0) 10 This value Dif corresponds to the distortion of the watermark inserted at the adder 52 generated in the quantization process.
On the basis of the absolute value Dif and the sign S watermark correction signals are generated in step 506.
15 The process in step 506 is described later. After the generation of the watermark correction signals for the DCT coefficients related to the k-th element of the watermark, index k is compared with N in step 508. If k > N, then the process is finished. If k <_ N, the value 20 k is increased by one in step 510, and the process goes back to step 502. The watermark correction process is thus performed, and the obtained watermark correction signals are finally outputted to adder 64.
Next, we describe the process performed in step 506 25 in Figure 5, using the flow chart in Figure 6. In step ' '. 26 506, the watermark correction signals for the DCT related to the k-th element of the watermark are generated. The array ~Dq k[i] (i=1,...,n k) are first all set to zero in step 520. Next, the value j is set to one in step 522.
Then the index i of the DCT coefficients is found through a permuting function p(j) in step 523. The function p(j) returns the j-th value of a permutation obtained after the integers 1 to n k are permuted. The simplest example is p(j)=j. Next, a value -s is stored in ~Dq k[i], and the value Dif is decreased by Q k[i]F k[i] in step 524.
This indicates that the quantization value Dq-k[i] is changed by -s by adding ~Dq k[i] to Dq k[i] at adder 64.
The value s is 1 or -1, so the change in the quantization value is one. In other words, de-quantized value obtained in a inverse quantizer in a decoder is changed by -Q k[i], one step size. The value Dif after the update is identical to the absolute value of the difference between cu[k] and Vr[k] calculated with the corrected quantization values Dq k[i] + ~Dq k[i] (i=1,...,n-k). After step 524, the values Dif and zero, and index j and n k are compared in step 526. If Dif < 0 or j>n k then this subroutine is finished. If the condition is not satisfied, the index j is increased by certain amount ~j in step 528, then the process returns to step 523.
Instead of the process shown in Figure 6, an alternative process shown in Figure 7 can be used as the ' '. 27 process of step 506. In the process shown in Figure 7, a step 530, checking whether the quantization value Dq-k[i]
equals zero or not, is added between step 523 and step 524. In this case, step 524 is performed only if the quantization value Dq k[i) is a non-zero value. This allows a reduction in the increase of the number of bits caused by correcting watermark information, because changing a quantization value from zero to non-zero value generally results in a large increase in the number of bits.
Figure 4 is a schematic diagram of an alternative embodiment of a modified MPEG-2 encoder for reducing degradation of a watermark in watermarked data. In this embodiment, the basic concept is the same as the described in connection with Figure 3. The differences lie in the fact that subtraction of the predicted images from the input original images is performed in the DCT
domain not in the spatial domain. For the predicted image outputted from the frame memory 32, DCT is performed in DCT calculator 70, and the results are subtracted from the watermarked DCT coefficients at subtractor 72. The results of subtraction are sent to the quantizer 36 and then the watermark correction is performed in the same manner as shown in Figure 3. The results outputted from subtractor 72 are the same as the results outputted from the DCT calculator 34 in Figure 3 because of the linearity of DCT. Therefore, the results obtained in the ' '. 2 8 processes followed by quantization in quantizer 36 are the same as those in Figure 3. This embodiment results in a reduction in the number of DCT calculations.
The detection procedure to detect a watermark in an image will now be described.
If MPEG video is the input image data format, the following detection process determines whether watermark W is present, where W[1,...,N) is the watermark being tested for.
Decode the Huffman code, but do not compute the inverse DCT's, so that, for each frame (at least, each I-frame), there is an array of 8x8 DCTs.
Next perform the same summation of DCT coefficients that was performed during watermark insertion to obtain the vector V. Compute the correlation coefficient C, between v and the watermark being tested for, W:
vl=v-v w'=w-w _ Finally, convert C into a normalized Fisher Z
statistic:
2o Z= N-31o 1+C
2 ~ 1-C
where N is the length of the watermark.
The Z value indicates whether the watermark is present. A preferred threshold for Z is 4 (i.e. Z >_ 4 means the watermark is present), but other values may be used depending on the desired probabilities of false alarms and missed detec:tions.
Figure 8 is a flow diagram of the detection method for MPEG video input described above. The input MPEG
video is subject to a Huffman decoder and partial parser 80 where the output is a set of DCT for nxn, preferably 8x8, blocks of the video input.
The nxn DCTs are provided to watermark accumulator 82. Accumulator 82 has memory whose length is the watermark length N. DCT coefficients from the Huffman decoder and partial parser 80 are classified. according to a predetermined rule and summed for extracting a watermark as mentioned before, and the results are accumulated in the memory. The extracted watermark is provided to comparator 84 where it is compared with possible watermarks in the image by calculating correlation coefficients between the extracted watermark and the possible watermarks as mentioned before. The possible watermarks are the set of the watermarks that might have been inserted into the video data. The details of the accumulators and comparator are found, for instance, in U.S. Patent 5,915,027 issued June 22, 1999.
The output of cornparator 84 is the li~;elihood (normalized Fisher Z statistic) of the detected watermark being each of the possible watermarks. The most likely watermark is determined and is deemed the watermark in the image, or, if the detector does not exceed a predetermined threshold, then no watermark is present.
Alternatively, if the incoming input data comprises 5 an uncompressed image, an embedded watermark can be detected by applying the method above to DCT coefficients obtained by performing 8x8 DCT for the whole image. In this case, DCT have to be performed for each 8x8 block, but a skillfully designed rule for classifying DCT
10 coefficients into N sets enables us to avoid performing DCT many times. Before explaining the method to reduce DCT calculation, let us define some notations.
Let h m(i,j) (i=1,...,8, j=1,...,8, m=0,...,M-1) be a set of functions that map frequency indices of 8x8 DCT
15 coefficients (i, j) onto the indices k of the element of a watermark, and M indicates the number of functions. So if k=h m(i,j), then a DCT coefficient whose index is (i,j) is classified into the Ic-th set of DCT coefficients .
for calculating the value V[k]. We prepare M different 20 functions h m(i,j) (m=0,...,M-1). Which function is selected for a certain 8x8 block depends on the numbers r and c of the block where r and c indicate the row and column numbers of the block respectively. So the index of the functions, m, is first found according to the values 25 r and c, then the index of the sets, k, is determined by h m(i,j) for each DCT coefficient. Using these functions h m(i,j) in classifying DCT coefficients is assumed in the remaining part of the detailed description.
In this case, we can reduce the number of DCT
calculations in the following manner. First the sum of the blocks whose indices m are the same is computed for each m=0,...,M-1. Let this summed block be denoted. by VB m[i,j] (m=0,...,M-1). Then DCT is perfomred for the M
summed blocks VB m[i,j] (m=0,...,M-1). Fina7.ly, the DCT
coefficients of the summed blocks are classi:Eied into N
sets according to the value h m(i,j), and added together within each set to obtain V[1,...,N]. The obtained results V[1,...,N] are the same as V[1,...,N] obtained with the method mentioned above because DCT .is a linear transform, that is, the sum of DCT blocks equals the result of DCT for the sum of the blocks. If M is much less than the total number of blocks in an image, the number of DCT calculations is dramatic ally reduced. This method thus allows us to extract watermarks with small calculation cost.
Figure 9 shows a flow diagram of the d~=_tection method for uncompressed video input data as described above.
The uncompressed video image data is provided to nxn accumulators, preferably 8x8 accumulators 90. The memory requirement is n2 times the number of the function h m(i,j). For each index m, the blocks with the same index m are summed and the resultant M summed blocks are accumulated in the memory.
The output is the summed signal of each of the nxn blocks. The output is subject to a DCT transform 92. The number of transformations is proportional to the number of the functions h m(i,j). The result is a group of nxn DCTs which is classified into N sets according to the functions h m(i,j) and summed for extracting a watermark as mentioned before. The obtained watermark is provided to watermark accumulator 94 and accumulated. The memory requirement for accumulator 94 is proportional to the , watermark length. The extracted watermark is provided as input to comparator 96. The other inputs to comparator 96 are the possible watermarks that may have been inserted into the input image data. The comparator computes a likelihood (normalized Fisher Z statistic) of each possible watermark having been inserted into the image data. The most likely watermark is determined and is deemed the watermark in the image.
A limitation of block based DCT methods is their sensitivity to spatial shifts of the image. For example, if the image is shifted two pixels to the right, then the DCT coefficients change significantly, so that the watermark cannot be detected. Furthermore, general distortions, such as scaling and rotation, also make the watermark undetectable.
To solve these problems, the above insertion and extraction methods may be modified in two ways. The first possible modification is to insert multiple watermarks designed to survive predefined distortions of the video.
The second modification is to arrange that translations can be compensated for without performing the summation more than once. Optionally, this second modification may be further modified to insert registration patterns, one for each of the multiple watermarks, which can be used by a modified watermark detector to compensate for arbitrary translations of the video.
When detecting watermarks inserted using the above method, it is necessary to divide the image into the same grid of nxn blocks as was used during insertion. If the image has been translated since watermark insertion, then determining the correct grid becomes difficult. In many °
applications, this is a serious problem, since certain, specific transformations can be expected. For example, it can be expected that video on a DVD disk might be modified to fit on a standard television screen by conversion to either "panscan" or "letterbox" mode. In "panscan" mode, the horizontal image resolution is increased, and the image is cropped at a predetermined offset, so that the resulting image will be correct when viewed on a 3x4-aspect-ratio television screen. In "letterbox" mode, the image is scaled vertically, and black is added at the top and bottom, so that the whole image will fit correctly on a 3x4-aspect-ratio screen.
Since these two geometric transformations are more likely than any other, it is reasonable to prepare for them specifically.
The problem of the predetermined scaling or transforming of watermarked video is solved in the present invention by inserting an additional watermark for each of the likely transformations. Each of these watermarks is designed so that, when the image has undergone the corresponding known transformation, the grid of nxn blocks used during insertion will align with a predetermined grid used during detection. Thus, if the image has undergone no transformation, then the detection grid will align with the normal mode watermark, and the normal mode watermark will be detected. If the image has undergone "panscan" transformation, then the same detection grid will align with the "panscan" watermark, and the "panscan" watermark will be detected, and so forth for "letterbox" scan or any other predefined transformation.
The procedure for inserting a watermark that is to be detected after a specific transformation comprises the following steps:
1. Make a copy, IT, of the image being watermarked, I, and apply the transformation to be compensated for the image. For example, IT might be a copy of I that has been transformed into "letterbox" mode by vertical shrinking of I.
2. Create a watermarked version of the transformed image, IT', according to the general watermarking method described above.
3. Let WT = IT'-~IT be the spatial pattern that was added to IT when it was watermarked.
5 4. Perform the inverse transformation on WT to yield the corresponding watermark pattern, W, for the untransformed image. ror example, if the transformation to be compensated for was "letterbox" mode, then W would be obtained by vertically expanding WT.
10 5. Let I' - I + W be the image with a watermark added for the given transformation.
When the transformation is applied to the watermarked image I', the result will be approximately IT', 15 and the watermark will be detected by the same procedure designed to detect a normal watermark.
This process can only be used for a small number of transformations, as each additional watermark causes additional degradation of the image, and reduces the 20 detectability of other watermarks in the image. However, tests have shown that three watermarks --two for transformed images and one for the untransformed image--result in acceptable fidelity and good detectability.
Alternatively, for video, each watermark ca:n be inserted 25 in a time-multiplexed manner.
In cases where the transformations that an image will undergo are not predefined or predetermined, or where there are too many probable transformations to allow for the insertion of a separate watermark for each transformation, the above described method of compensating for transformations is not optimal. Thus, the present invention includes an additional improvement, which compensates for arbitrary translation of the image between the times of watermark insertion and watermark detection.
Arbitrary translation is compensated for in two ways. One way is by translations by even multiples of 8 pixels in the x or y directions when 8x8 blocks are used.
This can be easily compensated for if the following restrictions are imposed on the relationship between an index of function h m(i,j), m, and the row and column numbers of blocks, r and c. We determine the index m according to m = f(r,c) mod M, where f(r,c) is a linear function of r and c, and f(0,1) and f(1,0) are integers.
In addition, let h m(i,j) be a function expressed as h m(i,j) - (h-0(i,j) + axm) mod N, where a is an integer. These restrictions are assumed in the remaining part of the detailed description. In this case, the shift compensation is performed in the basic detection algorithm by computing the correlation of the extracted watermark with all cyclical shifts of the watermark being tested for. This is because the values VB m[i,j] (m=0,...,M-1) obtained from a watermarked image shifted by a multiple of 8 pixels in the horizontal and/or the vertical directions are identical to the values VB m[i,j] (m=0,...,M-1) obtained from the non-shifted image except that the indices m are cyclically shifted. As a result, the extracted value V[1,...,N]
obtained from a watermarked image shifted by a multiple of 8 are identical to the value V[1,...,N] obtained from the non-shifted image except that the elements V[k] are cyclically shifted. For nxn grid format, the same thing is true of a shift of a multiple of n in the horizontal and/or the vertical directions.
For shifts of less than 8 pixels in x and/or y directions, an exhaustive search can be performed of all 64 possibilities and the maximum Z value taken from the set of 64xM - Z values, where M is the number of 8x8 accumulators. In our tests M was chosen to be 64. The factor M is necessary to account for the cyclic shifts that are introduced by shifts of a multiple of 8 pixels.
The exhaustive search requires shifting the M 8x8 accumulator array in the spatial domain and then performing the DCT of each of the M 8x8 blocks. This is performed 64 times, once for each of the possible shifts.
Thus it is necessary to perform 64xM 8x8 DCTs. If this computation is too expensive in terms of time or memory an alternative method can be used, as described below.
The second way, which compensates for translations of non-even multiples of n pixels, uses a pattern (referred to as a "registration pattern") which can be inserted at the time of watermark insertion. By finding the location where the registration signal best matches a predefined signal, a detector can determine how much to shift the data before extracting the watermark. This shifting must be done in the spatial domain, but can be done with accumulators, so conversions of whole images are avoided.
Moreover, the 64xM 8x8 DCTs are unnecessary.
Instead, the correct registration is determined in the spatial domain and then compensated for by shifting the pixels in the accumulator arrays. The M 8x8 accumulators are only then transformed into the DCT domain and the watermark extraction is performed as described above.
The registration pattern is an 8x8 spatial pattern inserted into the image in such a way that the sum of all 8x8 pixel blocks highly correlates with the pattern.
Again, an nxn spatial pattern is used when the video is nxn blocks. A registration pattern can be inserted by using the watermark insertion method described above. The sum of all 8x8 pixel blocks becomes highly correlated with a registration pattern if the DCT coefficients of the sum block and those of the registration pattern are highly correlated to each other. In addition, the DCT
coefficients of the sum block equals the sum of all DCT
blocks because of the linearity of DCT. We can thus insert a registration pattern using a method similar to that inserting a watermark described above, considering the DCT coefficients of a registration pattern as a watermark W and considering the sum of all 8x8 DCT blocks as a value V.
Insertion is performed by converting the registration pattern into the DCT domain, and then using the basic insertion algorithm with different DCT
coefficients Dks and filter coefficients Fks.
Specifically, each AC term of the registration pattern's DCT is considered one element of the watermark, W[k]. The set of DCT terms that are summed together to extract this element, Dk, is simply the set of corresponding terms of all the 8x8 DCTs in the image. All the Fks are set to 1. Using these Dks and Fks, the insertion algorithm inserts a registration pattern along with each watermark. The watermarks are still inserted with the original Dks and Fks.
During detection, registration is performed as follows. Each of a predetermined number of blocks is summed together to form a single nxn, typically 8x8, block.
We have arbitrarily used 64 blocks in our tests. This block contains a registration pattern placed there by the insertion process. To determine the horizontal and vertical translation of the frame, a correlation process is performed in the spatial domain to determine these offsets. Sixty-four correlations are performed for each of the 8 horizontal and 8 vertical motions that are 5 possible. When the 8x8 patch is shifted either horizontally or vertically, a wrap around shift is performed.
We now describe how to determine the shift of the grid. In the following method, we assume the integer 10 values f(0,1) and f(1,0) are relatively prime to M.
First, the blocks for which the same function h m(i,j) is used in classifying DCT coefficients are added together in the spatial domain for generating a summed block VB m[i,j] for each m. The sum of all nxn 15 blocks denoted by AB[i,j], is computed by !1T-1 ~B~l,,j] - ~vB-1111, J~ ~1 - x,...,11, J = 1,...,11.
m=0 Then the correlation coefficient between AB[i,j]
and a registration pattern, denoted by R[i,j], is computed. After calculation of the correlation 20 coefficient, the values AB[i,j] are cyclically shifted by one column in the horizontal direction, and the correlation coefficient between AB[i,j] and R[i,j] is calculated in the same way. The same operations are repeated for each shift. After shifting n times, AB[i,j]
25 becomes identical to AB[i,j] before any shift is done.
Then, the values AB[i,j] are cyclically shifted by one row in the vertical direction, and calculation of a correlation coefficient and shift by one column in the horizontal direction are repeated. In this way, we can calculate correlation coefficients for all n2 possible shifts. At the same time, we search the shift value (offset), denoted by (X, Y), which gives the maximum correlation coefficient.
After the offset (X,Y) has been determined, the M
summed blocks VB m[i,j] are then shifted accordingly in the spatial domain.
Next, we describe the method to compensate for the shift value X in the horizontal direction. To do so, the values VB m[i,j] (m=0,...,M-1) are first copied on an nxnM array VB1[i,j] (i=1,...,n, j=1,...,nM). In copying the values VB m[i,j], the spatial relationships between blocks, that is, which blocks are adjoining a certain block are considered. The function f(r,c) is linear, so f(r,c+1) - f(r,c) + f(0,1).
This indicates that the blocks whose index m equals ml = (m0 + f(0,1)) mod M
are located next to the blocks in which m=m0. So, the values VB ml are copied next to the values VB m0 in the array VB1. For this reason, for each I (I=0,...,M-1), the values VB m'(i,j] (i=1,...,n, j=1,...,n, m'=Ixf(0,1) mod M) are copied in the nxn square region of VB1 where the index of the left-top corner is (1, nI+1). After copying the values for all I, the array VB1 is filled with the values VB m[i,j] because f(0,1) and M are relatively prime. Next, the values in the array VB1 are cyclically shifted by X in the horizontal direction. Then the values in VB1 are returned to VB m in such a way that the value in the nxn region of VB1 where the index of the left-top corner is (1, nI+1) are substituted to VB m'[i,j]
(i=1,...,n, j=1,...,n, m'=Ixf(0,1) mod M) for each I
(I=0,...,M-1). The horizontal offset value X can thus be compensated for without shifting the whole image itself.
Next, the shift value Y in the vertical direction is compensated for. To do so, the values VB m[i,j]
(m=0,...,M-1) are first copied on an nMxn array VB2[i,j]
(i=1,...,nM, j=1,...,n). In copying the values VB m[i,j], the spatial relationship between blocks are considered.
The function f(r,c) is linear, so ' f (r+l, c) - f(r, c) + f (1,0).
This indicates that the blocks whose index m equals ml = (m0 + f(1,0)) mod M
locate under the blocks in which m=m0. So, the values VB ml are copied under the values VB m0 in the array VB2.
For this reason, for each I (I=0,...,M-1), the values VB m'[i,j] (i=1,...,n, j=1,...,n, m'=Ixf(0,1) mod M) are copied on the nxn square region of VB2 where the index of the left-top corner is (nI+1, 1). After copying the values for all I, the array VB2 is filled with the values VB m[i,j] because f(1,0) and M are relatively prime. Next, the values in the array VB2 are cyclically shifted by Y
in the vertical direction. Then the values in VB2 are returned to VB m in such a way that the value in the nxn region of VB2 where the index of the left-top corner is (nI+1, 1) are substituted in VB m'[i,j] (i=1,...n, j=1,...,n, m'=Ixf(0,1) mod M) for each I (I=0,...,M-1).
The vertical offset value Y can thus be compensated without shifting the whole image itself.
The offset of the nxn grid is compensated by the process mentioned above. These processes are performed in the registration process 108 (Figure 10) as will be explained later. A shift of a multiple of n remains even after these processes, but this shift does not affect the watermark detection because the correlation coefficient between a watermark W and an extracted value V is calculated by shifting the watermark W cyclically as described above. After the registration process above is applied, the M blocks VB m (m=0,...,M-1) are transformed back to the DCT domain.
With reference now to Figures 10 and 11, there are shown the basic detection algorithms modified to compensate for translational registration. In the case of MPEG video input (Figure 10), 8x8 DCT blocks obtained from an MPEG video stream are first classified into M
groups according to their indices m of the function h m(i,j), summed within the groups for generating M
summed blocks, and the resultant summed blocks are accumulated in 8x8 accumulators 102. The M summed blocks in accumulators 102 must be converted into the spatial domain by performing an inverse DCT operation in inverse DCT converter 104, and accumulated in accumulators 106.
Finding the offset value of the 8x8 grid and compensating for the offset is executed for the output from 8x8 ' accumulators 106 in registration 108 as described above.
The registration data outputted from registration process 108 is accumulated in accumulators 110 and converted into the DCT domain in DCT converter 112 for watermark extraction by use of accumulators 114, watermark extractor 116 and watermark decoder 118. In watermark extractor 116, the DCT coefficients outputted from accumulator 114 are classified into N sets according to the functions h m(i,j) and summed for extracting a watermark. The obtained watermark is provided to watermark decoder 118, in which the processes executed in comparator 84 in Figure 8 for finding a watermark corresponding to the extracted watermark. The watermark considered to have been inserted is outputted from the watermark decoder 118. In the case of uncompressed input data (Figure 11), the input data is divided into 8x8 blocks and accumulated in accumulators 106 according to the indices of the functions h m(i,j), and registration 108 is performed before conversion into the DCT domain in DCT converter 112. The process continues as described above.
While there has been described and illustrated methods of insertion and detection of watermarks in image 5 data, it will be understood by those skilled in the art that variations and modifications are possible without deviating from the spirit and broad teachings of the present invention which shall be limited solely by the scope of the claims appended hereto.
U.S. Patent 6,069,914 issued May 30, 2000, entitled "Watermarking of Image Data Using MPEG/JPEG Coefficients", advances this work by using MPEG/JPEG coefficients to encode the image data.
U.S. Patent 5,91~i,027 issued June 22, :L999, entitled "Digital Watermarking", describes storing watermark information into subimages and extracting watermark information from subimages.
A review of watermarking is found in an article by Cox et al., entitled "1?, review of watermarking and the importance of perceptual modeling" in Proc. of EI'97, vol.
30-16, February 9-14, 1997.
There have been several proposals to watermark MPEG
video or JPEG compressed still images. In a1.1 cases, each 8x8 DCT block is modified to contain the watE~rmark or a portion thereof. Consequently, decoding of the watermark requires that each 8x8 block be individually analyzed to extract the watermark signal contained therein. The individual extracted signals may then be combined to form a composite watermark, which is then compared with known ' ' 7 watermarks. Because each block must be analyzed , individually, an uncompressed image must be converted back to the block-based DCT representation, which is computationally expensive. Thus, while the decoder may be computationally efficient in the DCT domain, extracting a watermark from the spatial domain is much more expensive.
To allow for computationally efficient detection of the watermark in both the spatial and DCT domains, a watermark may be inserted in the sum of all the 8x8 blocks in the DCT domain, or the sum of a subset of all the 8x8 blocks in the DCT domain. A major advantage of this approach is that if the image is only available in the spatial domain, then the summation can also be performed in the spatial domain to compute a small set of summed 8x8 blocks and only those blocks must then be transformed into the DCT domain. This is because the sum of the DCT blocks is equal to the DCT of the sum of the intensities. Thus, the computational cost of decoding in the DCT and spatial domains is approximately the same.
A second advantage of watermarking the sum of the DCT blocks is that there are an unlimited number of equivalent methods to apportion the watermark throughout the image. For example, if the watermark requires a change of 0i to the i-th coefficient of the summed DCT
block, then, if there are M blocks in the image, ~i/M can be added to each individual block, or block 1 can have ~i added to it and the remaining M-1 blocks left unaltered, ignoring for the moment issues of image fidelity. Because of this one to many mapping, it is possible to alter the insertion algorithm without changing the decoder. This is a very important characteristic, since in some watermarking applications, there may be many hardware decoders that are deployed, such that changing the decoder is impractical. However, improvements to the insertion algorithm can still result in improved detection using the approach described herein.
A third advantage of watermarking the sum of the DCT blocks is that watermark signals extracted from these sums have small variances, compared with the amount that they may be changed without causing fidelity problems.
This means that, in many cases, it is possible to change an image so that the summed DCT blocks perfectly match the required watermark signal, even though the resulting image appears identical to the original.
Finally, it is well known that some problems, such as modeling the human visual system, are best performed in the frequency domain, where other problems such as geometric transformations are more conveniently dealt with in the spatial domain. Since the computational cost of decoding the watermark is now symmetric, it is possible to switch from spatial to frequency domains at will in order to correct for various signal transformations that may corrupt the watermark.
Summary of the Invention In accordance with the present invention, there is provided a method of inserting a watermark signal into data to be watermarked comprising the steps of: applying a watermark extraction to unwatermarked data for generating extracted values; computing a difference between a watermark to be inserted and the ~~xtracted values; and adding the computed difference throughout the unwatermarked data.
In accordance with the present invention, there is also provided a method of detecting a watermark from watermarked data comprising the steps of: receiving 'watermarked data where the watermarked data comprises nxn :blocks of watermarked data; summing the nxn blocks of watermarked data to form at least one nxn blo~~k of summed watermarked data; cyclically shifting the blo~~k in a spatial domain to register the block; and extracting 'the watermark from the at least one n:~n block.
In accordance with the present invention, there is <also provided a method of detecting a watermark from watermarked data comprising the steps of: receiving watermarked data where t=he watermarked data comprises nxn blocks of watermarked data; summing the nxn blocks of watermarked data to forrn at least one nxn block of summed watermarked data; and extracting the watermarlt from the at :Least one nxn block by summing predetermined watermarked data from each nxn block to yield each elements of the extracted watermark.
In accordance with the present invention, there is also provided a method of detecting a waterma:rk from watermarked data comprising the steps of: receiving 9a watermarked data where the watermarked data comprises nxn blocks of watermarked data; summing the nxn blocks of watermarked data to form M nxn blocks of summed watermarked data, where M is number of functions mapping frequency indices onto indices of elements of watermarks; and extracting the watermark from the at least one nxn block.
In accordance with the present invention, there is also provided a method of inserting a watermark signal into video images comprising the steps of: receiving input video images; performing discrete cosine transformation (DCT) of said input video images to obtain input video image DCT
'values; adding watermark signals to said input video image DCT values to obtain watermarked DCT values; performing an inverse DCT on the watermarked DCT values for generating watermarked images; subtracting predicted ima!~es stored in memory from said watermarked images for generating residual images; performing discrete cosine transformation of said residual images to obtain residual image DCT values and quantizing the residual image DCT values to d~°rive residual image quantized DCT values; inverse quantizing the residual :image quantized DCT values to obtain residual image de-quantized DCT values; performing discrete cosine transformation of the predicted images to obtain predicted :image DCT values; summing the residual image de-quantized DCT values and the pred_~cted image DCT values for generating decoded image DCT coefficients; calculating correction signals by applying a watermark extraction to the decoded :image DCT coefficients and comparing resultani~ extracted values with watermark target values; adding the correction signals to the residual image quantized DCT values for obtaining an output signal; inverse quantizing the output signal to obtain inverse quantized output data, performing inverse DCT of the inverse quantized output signal and 9b summing the resultant signal with the predicted images for generating a summed signal; generating predicted images to be stored in said memory by using the summed signal; and variable length encoding the output signal for providing a watermarked MPEG video signal of the input video images.
In accordance with the present invention, there is also provided a method of inserting a watermark signal into video images comprising the steps of: receiving input video images; performing discrete cosine transformation (DCT) of said input video images to obtain input video image DCT
values; adding watermark signals to said input video image DCT values to obtain watermarked DCT values; performing discrete cosine transformation on predicted images stored in memory to obtain predicted image DCT values; subtracting the predicted image DCT values from the watermarked DCT values to obtain residual image DCT values; quantizing the residual image DCT values to obtain residual image quantized DCT
values; inverse quantizing the residual image quantized DCT
values to obtain residual image de-quantized DCT values and summing the residual image de-quantized DCT values with the predicted image DCT values, for generating decoded image DCT
coefficients; calculating correction signals by applying a watermark extraction to the decoded image DCT coefficients and comparing resultant extracted values with watermark target values; summing the correction signals with the residual image quantized DCT values to obtain an output signal; inverse quantizi.ng the output signal to obtain inverse quantized output. signal, performing inverse DCT of t:he inverse quantized output signal and summing the resultant signal with tree predicted images to provide locally-decoded images; generating predicted images to be stored in said memory using the locally-decoded images; and 9c valuable length encoding the output signal to provide a watermarked MPEG video signal of the input video images.
In accordance with the present invention, there is also provided a method ~f inserting an extractable watermark signal into image data so that the watermark signal can be extracted after the watermarked data is subject to distortion comprising the steps of: providing image data to be watermarked; copying the image data; applying a predefined distortion to the copy of image data; inserting a watermark signal into the distorted copy of image data;
subtracting the copy of distored image data from the watermarked copy of distorted image data to generate a watermark pattern; performing inverse distortion on the watermark pattern to generate the watermarked pattern for nondistorted image data; and combining said image data to be watermarked and said watermark pattern for nondistorted image data to obtain watermarked data.
In accordance with the present invention, there is also provided a method of inserting an extractable watermark signal into image data so that the watermark ;signal can be extracted after the watermarked data is subject to distortion comprising the steps of: receiving image data in form of nxn blocks; and inserting an nxn registration pattern into the image data.
In accordance with the present invention, there is also provided a method of detecting a watermark from watermarked image data that has been subjected to a two-dimensional translation distortion comprising the steps of:
partitioning received distorted watermarked image data into a set of nxn blocks; summing together watermarked image data in a predetermined number of nxn blocks; if summed nxn blocks are not in spatial domain, transforming summed nxn 9d blocks into spatial domain; determining two-dimensional translation distortion of the summed nxn. blocks; shifting the summed nxn blocks according to the determined distortion; transforming the shifted nxn blocks into transform domain; extracting a possible watermark signal from the transformed nxn blocks; correlating the possible watermark with a known watermark; repeating said correlating for cyclic shifts of the possible watermark; and determining the presence of a watermark based on said correlating.
In accordance with the present invention, there is also provided a method of detecting a watermark from watermarked image data that has been subjected to a two-dimensional translation distortion comprising the steps of:
(a) partitioning received distorted watermarked image data :in nxn blocks; (b) summ_Lng together watermarked image data in a predetermined number of nxn blocks; (c) if summed nxn blocks are not in spatial domain, transforming summed nxn blocks into spatial domain; (d) shifting the summed nxn blocks by one of n2 possible shifts; (e) tran~;forming the shifted nxn blocks into transform domain; (f) extracting a possible watermark signal from the transformed nxn blocks;
(g) correlating the possible watermark with a known watermark; (h) repeating said correlating for cyclic shifts of the possible watermax-k; (i) repeating steps (d) - (h) for each of the n2 possible shifts; and (j) determining the presence of a watermark based on the results of step (i).
~e The present invention concerns a novel insertion method which employs a specific model of the human visual system which provides much better control over image fidelity. Tests have shown that it is possible to obtain large signals (more than 15 standard deviations fzom 0 correlation) with images that are indistinguishable from their respective original images.
The method handles robustness against various types of attacks in ways that are easy to relate to the specific type of attack.
The method is adaptable so that the model of the human visual system and the techniques used for handling attacks can be changed later without having to change the detector. The result is that it is possible to continue improving watermarking, particularly DVD (digital video disk) watermarking, even after many detectors have been installed. This is analogous to the situation with MPEG
video for which encoder technology can be improved without having to change existing decoders.
Use of the present insertion method allows a simple detection algorithm in either MPEG or decompressed domains.
The invention also concerns a novel detection method which is easy to implement, easy to analyze and has a low computational cost, whether the incoming video is MPEG compressed or uncompressed.
The present invention also concerns a novel insertion method that hides multiple patterns in the data.
These patterns fall into two categories: 1) registration 5 patterns used during detection to compensate for translational shifts, and 2) watermark patterns that encode the information content of the watermark.
A principal object of the present invention is the provision of a digital watermark insertion method which 10 allows detection of watermarks after the watermarked data is subjected to predefined scale changes, without modification to the watermark detector.
Another object of the invention is the provision of a watermark detection method that is computationally inexpensive in either the MPEG or decompressed domains.
A still other object of the invention is the provision of a digital watermarking method that withstands attacks without having to change a detector.
Further and still other objects of the invention will become more clearly apparent when the following description is read in conjunction with the accompanying drawings.
Brief Description of the Drawings Figure 1 is a general data flow diagram of a method of inserting a watermark into media data;
Figure 2 is a schematic diagram of an MPEG-2 encoder;
' '. 11 Figure 3 is a schematic diagram of a modified MPEG-2 encoder for reducing degradation of a watermark in watermarked data;
Figure 4 is a schematic diagram of an alternative modified MPEG-2 encoder for reducing degradation of a watermark in watermarked data;
Figure 5 is a flow chart of the process performed in the watermark correction device in Figures 3 and 4;
Figure 6 is a flow chart of the process performed in step 506 of Figure 5;
Figure 7 is a flow chart of an alternative process performed in step 506 of Figure 5;
Figure 8 is a flow diagram of a method of extracting a watermark from MPEG media data;
Figure 9 is a flow diagram of a method of extracting a watermark from uncompressed media;
Figure 10 is a flow diagram of a method of detecting a watermark from MPEG data, with registration;
and Figure 11 is a flow diagram of a method of detecting a watermark from uncompressed image data, with registration.
Detailed Description of the Preferred Embodiment As used in the following description the terms image and image data will be understood to be equally applicable to video, image and multimedia data. The term "watermark" will be understood to include embedded data, symbols, images, instructions or any other identifying information.
In order to better understand the present invention, first a review of the basic watermarking method will be presented followed by additional descriptions of the improvements comprising the present invention.
First, we define some notations. Let a watermark to be embedded into an image be an N dimensional vector, denoted by W[1,...,N]. In the following text, the notation W[1,...,N] is used in the same manner as W[k]
(k=1,...,N). Let V[1,...,N] denote a vector value extracted from an image, where the element V[k]
corresponds to W[k]. Specifically, the value V[k] is a weighted sum of DCT coefficients given by V[k]=D k[1]F k[1]+D k[2]F-k[2]+...+D k[n k]F k[n k], where D k[i] (i=1,...,n_k) indicate members of the set of DCT coefficients used for calculating V[k], n k indicates the number of members, and F k[i] (i=1,...,n k) are weighting coefficients related to a filter processing.
The concept of F k is that the DCT coefficients are weighted according to how much noise might be expected in each coefficient. To calculate V[1,...,N], nxn DCT
coefficients are first calculated over a whole image.
Then the coefficients are classified into N sets, each of which is related to each element of V[1,...,N]. The rule of classifying DCT coefficients is predetermined, and the same rule is used in both inserting and detecting a watermark.
Before inserting a watermark into an image, a detection algorithm is applied to the image to find a watermark that is already present in the image. If the image does not contain a watermark, the extracted values, V[1,...,N] will be normally distributed random numbers °
that do not correlate any watermark W[1,...,N]. A
watermark W[1,...,N] is inserted into an image by changing each of D k[1,...,n k] (k-i,...,N) slightly in order to make the extracted value V[1,..., N] highly correlate the watermark W[1,...,N]. Let the target value of V[1,...,N] be denoted by w[1,...,N], that is, the values V[1,...,N] are changed to.w[1,...,N] by inserting the watermark W[1,...,N]. The target values w[1,...,N]
have high correlation with the watermark W[1,...,N] and they are determined as will be described below.
After the target values ~[1,...,N] are determined, the difference w[k]-V[k] is distributed among the DCT
coefficients D k[1,...,n k]. Then a watermark is inserted by adding the allocated difference value to the corresponding DCT coefficients D k[1,...,n k]. This change of DCT coefficients must be done in such a manner so as not to change the appearance of the image.
In distributing the difference w[k]-V[k], a characteristic of the human visual process is taken into ' '. 14 account. The amount of change that does not cause a visible change in the image is different for each DCT
coefficient D k[i]. This amount depends on the human visual process which can be approximately simulated with a computational model. The amount of change is referred to as "slack". The slack is calculated for each DCT
coefficient and it is used in distributing the difference w[k]-V[k] among the DCT coefficients. Next we describe how to calculate the values of the slack using a model of the human visual process. .
The preferred computational model of human visual sensitivity that is used in the present invention is found in an article by Andrew B. Watson, entitled "DCT
Quantization Matrices Usually Optimized for Indirect Images" in SPIE, vol. 1913 (1993), pp. 2U2-216. This model was applied to watermarking in an article by Christine I. Podilchuk and Wenjun Zeng entitled "Digital Image Watermarking Using Visual Models", Proc. of EI'97, vol. 3016, February 9-14, 1997. The current invention differs from that of Podilchuk and Zeng in (i) not requiring the original unwatermarked image at the decoder and (ii) not extracting the watermark from the individual 8x8 blocks, but from the sum of a set of 8x8 blocks.
Other computational models are also usable.
For each element of the image's block DCT, d[i,j], this model computes a value called the element's "slack", S[i,j], which indicates how much a particular d[i,j]
15 , value may be altered before such an alteration becomes visible. The value is computed in three steps. The first step models the contrast masking phenomenon of the human visual system and models the visual sensitivity at different frequencies and handles the difference between visual sensitivity to changes in different frequencies.
The second step models the luminance masking phenomenon of the human visual system and handles the fact that the visual system is more sensitive to changes in dark regions than to changes in bright regions. The third step handles the fact that the sensitivity to changes depends in part on the percentage that the frequency is changing (i.e. a DCT term with a small value in it may only change a little, while one with a larger value may change more).
The perceptual model makes use of a matrix of values that indicate the relative sensitivity of the human visual system to the different terms of a spatial 8x8 DCT. The formulae for computing this matrix are available in an article by Albert J. Ahumada Jr. and Heidi A. Peterson entitled "Luminance-Model-Based DCT
guantization for Color Image Compression", in SPIE, vol.
1666, (1992) pp. 365-374.
After computing the slacks for all the 8x8 DCT's in the image, a slack can be assigned to each D k[1,...,n k].
Call these slacks S k[1,...,n k]. It is now possible to distribute the changes in the v's over all the D k's with minimal visual impact. This is done according to the following formula:
v,_l~[1]-D-h[1]+(~[h1]-Y[k])*S l~[i] .
~S-l~[j]F-k[j]
where D' k[1,...,n k] are the modified 8x8 DCT
coefficients, and co[k] and V[k] are the k-th elements of cu and V, respectively. The effect of this formula is to distribute the desired change in a given element of the watermark vector (o~[k] - v[k]) over all the DCT
coefficients that are summed at calculation of V[k] to produce that element, proportionately according to those DCT coefficients' slacks. To illustrate, consider two simple examples: 1) If all the slacks are 0 except for slack S k[m], then the sum of all the values S k[j]F k[j]
is equal to S k[m]F k[m], and only D k[m] is changed. It is changed by the full value of (c~[k]-V[k]). 2) If all the slacks are equal and all the coefficients F k[i]=1, then each D k[i] is changed by the same amount.
After making these changes, convert all the 8x8 DCT's back into the spatial domain, and the result is a watermarked image. It is easy to show that the sum of all the D-k[i]F k[i]'s for a given k will equal ~[k]. The process of making this is referred to as "inserting Omega into the image". The watermark extracted from the resulting image, if the image has not been attacked, will be exactly w, not w plus noise.
There are two important issues remaining to be discussed. First, how to decide on w, and, second, how to make the watermark robust.
Previously, the equivalent of co was computed as:
~ = V + a*W
where a is a small constant, and W is a zero-mean watermark signal. It is possible to use the same formula here, but it is too limiting to result in the strongest possible watermark using the present invention. In practicing the invention, it is often possible to insert an w that has perfect correlation with the watermark, W, without causing any visible change in the image. The following formula is used:
w = mean(V) + (3*(v - mean(v)) + a*W
This result is a weighted sum of the watermark signal and the original, noise (image) signal. If (3 is set to 0, the result is an cu that perfectly correlates with W.
The signal to noise ratio (SNR) for an unattacked image will be:
SNR = a*std(W)/(3*std(V) where std(X) is the standard deviation of X.
There are many ways to choose a and (3 based on optimizations to maximize different criteria such as fidelity or robustness.
'. 18 At this point there is a complete method of inserting watermarks. The method contains explicit modeling of human vision, but it does not contain any explicit method of making the watermark robust. In fact, the method as described so far will try to put as much of the watermark as possible into the high frequencies since these frequencies have the largest slack, but this is a poor thing to do from the point of view of robustness.
To make the watermark robust against a given set of , attacks or signal degradations, it is first necessary to consider how those attacks affect the various terms of the 8x8 DCT's in the image. Then, terms that are affected by attacks or signal degradations in similar ways are grouped together, and watermarked as if the group of terms were a separate image.
The following is a simple example. Suppose there is only concern about two possible attacks: cropping 24 columns of pixels from the left side of the image, or cropping 24 columns of pixels from the right side of the image. This results in three groups of DCT terms: those that come from the 3 left-most columns of 8x8 DCT's blocks, those that come from the 3 right-most columns, and those that come from the rest of the image. All the terms in each of these groups either survives or is destroyed by any given attack together. If each group is watermarked as though it were a separate image, then the watermark from at least one group will generally survive ' '. 19 attack (assuming that the 24-column cropping attacks are the only attacks possible), and the watermark that is extracted will consist of the correct watermark, from that group, plus some noisy watermar)cs from groups that were damaged by the attack.
A more interesting example is low-pass and high-pass filtering attacks. It is possible to group all the low frequencies together into one group, and all the high frequencies into one or more other groups. If the predetermined rule for classifying DCT coefficients into the N sets is designed in such a way that each set has coefficients of many different frequencies, then the complete watermarlc can be inserted into each group. Then, if the high frequencies are removed, the watermark will still be detectable in the low frequencies, and vice-versa.
The more groups the terms are divided into, the more robust will be the watermark. There is a cost because it will become increasingly more difficult to distribute the changes without causing visible distortion.
It is important to note that the best balance can be achieved after the detectors are in wide use. It is possible to modify the insertion algorithm to make watermarks robust against a wide variety of attacks without having to change detection at all.
Presently only three groups are used. One group collects together most of the low frequencies. Each of '. 20 the other two groups represents one higher frequency.
In the preferred method, a watermark is not placed in any of the higher frequencies. The reason is not that they are susceptible to attack (that is handled by the design of the filter used in detection). Rather, it is because watermarking the higher frequencies causes MPEG
compression rates to go down substantially. An alternative solution might be to add other groups that contain higher frequencies.
In Figure 1, there is shown a flow diagram of the watermark insertion method. The digital image is divided into a collection of nxn blocks, preferably 8x8 blocks, in step 10. The discrete cosine transform (DCT) of each block is computed in a known manner in step 12. The DCT's are separated into groups that respond to different attacks in the same manner in step 14. A first group G is selected in step 16.
Next, extract a watermark V, using only the DCT
terms in the group G in step 18. Determine a new signal (target value) w selected such that cu is similar to V but is highly correlated with watermark W in step 20.
Add fractions of w-V to the DCT terms in G
according to perceptual slack in step 22. Decide whether group G is the last group in step 24. If not, select next group G in step 26, and extract watermark V using only terms of next group G in step 18 and continue procedures until the last group G is found in step 24. Next, compute ' 21 the inverse DCTs of the blocks in step 28 resulting in a watermarked image.
Some alternative steps in the insertion method are possible. For example, the distribution of the difference between cu and V over DCT terms can be done stochastically to help deter tampering and reduce susceptibility to tampering.
Also, the groups of DCT terms for robustness purposes could be performed dynamically. For example, the insertion program could simulate various attacks on the image and determine the effect on the values of the DCT
term. Then, the program would cause appropriate allocation of the terms into the groups similarly affected.
The inserter can be designed with a user-interface that allows the user to set two parameters: (1) the maximum perceptual difference (J) between the original image and the watermarked image and (2) the maximum allowable probability of missed detection after any of , the predefined set of attacks. The algorithm would then insert watermarks into a large number of images automatically, according to the allowable perceptual change J and checking each one against simulations of the attacks. If an image fails to meet the specified robustness constraint (maximum allowable probability of missed detection), then the user would be notified so that a manual decision can be made to compensate or trade-off image fidelity for robustness.
In addition, the distribution of the difference between w and V over DCT terms can be modified to explicitly compensate for MPEG quantization. Using the above watermark insertion method there may result a degraded watermark in the watermarked data. In order to enhance the watermark in the watermarked data after MPEG
compression several techniques are possible.
Figure 2 is a schematic diagram of a typical MPEG-2 encoder. Figure 2 depicts elements which are indispensable to execute an MPEG-2 encoding of P pictures, or to perform a combined interframe prediction and DCT
coding. Input images are provided as one input to subtractor 30. The other input to subtractor 30 is predicted image generated in frame memory 32. The predicted images are subtracted from the input images at subtractor 30. A discrete cosine transform (DCT) is performed at DCT calculator 34 on the output signal from subtractor 30. The DCT coefficients are quantized in quantizer 36. The outputs of the quantizer 36 are sent to a variable length encoder 38 where Huffman encoding is performed. The quantized DCT coefficients outputted from the quantizer 36 are also sent to an inverse quantizer 40 where they are de-quantized. Inverse DCT of the de-quantized DCT coefficients is performed in the inverse DCT calculator 42. The results are added at adder 44 to the predicted image outputted from the frame memory 32, and then an image which is expected to be the same as that acquired in a decoder is reconstructed. The reconstructed image is called "a locally decoded image."
This locally decoded image is stored in the frame memory 32 to produce the predicted images.
Figure 3 is a schematic diagram of a modified MPEG-2 encoder for reducing degradation of a watermark in watermarked data. Before MPEG-2 encoding, DCT is performed on a input image at the DCT calculator 50 and watermark signals are added to the DCT coefficients at adder 52. The output DCT coefficients including watermark information is subject to inverse DCT in the inverse DCT
calculator 54. The output of inverse DCT calculator 54 are images with a watermark. These watermarked images are sent to MPEG-2 encoder and MPEG-2 encoding is performed as described above. In addition, in this embodiment, watermark information is modified in order to be suited to MPEG-2 compression. DCT coefficients for the predicted images are calculated in DCT calculator 56. The quantization values outputted from quantizer 36 are de-quantized in inverse quantizer 58. The results of inverse quantization are added at adder 60 to the DCT
coefficients outputted from DCT calculator 56. The results of addition correspond to the DCT coefficients for the decoded images which are expected to be generated in a decoder. These DCT coefficients are inputted into a watermark correction device 62. The watermark correction '. 24 device 62 outputs watermark correction signals. At adder 64, the watermark correction signals from device 62 are added to the quantization values from quantizer 36. The output of adder 64 is used as the inputs to variable length encoder 38 and inverse quantizer 40.
Next, we describe the process performed in the watermark correction device 62. Let us introduce several new notations to explain the process. Let Dq k[i] be the quantization value corresponding to D k[i], that is, the quantization value of the i-th member of the k-th set for calculating the value V. Let Q k[i] be the quantization step size used in obtaining Dq k[i]. Let Dr k[i] be the output value of adder 60 that is obtained by adding the inverse quantization value of Dq k[i] calculated in inverse quantizer 58 to the corresponding DCT coefficient outputted from DCT calculator 56. Let Vr[1,...,N] be the value extracted from the output values of adder 60, Dr k[i], in the same manner that the value V[1,...,N] is calculated in inserting a watermark. We assume that the target value w[1...,N] is also available in the watermark correction device 62.
Figure 5 shows the flow chart describing the process performed in the watermark correction device 62.
First the index k of the watermark element is set to 1 at step 500. Next the value Vr[k] is calculated at step 502 by Vr[k] - F k[1]Dr k[1] +...+ F k[n_k]Dr k[n k], where the weighting coefficients F k[1,...,n k] are the same value as those used in calculating V[k]. Then the absolute value of the difference between the value Vr[k]
and the target value co[k], and the sign of the difference 5 are computed in step 504 by the following equations:
Dif = ~Vr[k]-w[k]~
s = Sign(Vr[k]-cu[k] ), where 1 ( x >_ 0) S1~'11 (X ) -1 (x<0) 10 This value Dif corresponds to the distortion of the watermark inserted at the adder 52 generated in the quantization process.
On the basis of the absolute value Dif and the sign S watermark correction signals are generated in step 506.
15 The process in step 506 is described later. After the generation of the watermark correction signals for the DCT coefficients related to the k-th element of the watermark, index k is compared with N in step 508. If k > N, then the process is finished. If k <_ N, the value 20 k is increased by one in step 510, and the process goes back to step 502. The watermark correction process is thus performed, and the obtained watermark correction signals are finally outputted to adder 64.
Next, we describe the process performed in step 506 25 in Figure 5, using the flow chart in Figure 6. In step ' '. 26 506, the watermark correction signals for the DCT related to the k-th element of the watermark are generated. The array ~Dq k[i] (i=1,...,n k) are first all set to zero in step 520. Next, the value j is set to one in step 522.
Then the index i of the DCT coefficients is found through a permuting function p(j) in step 523. The function p(j) returns the j-th value of a permutation obtained after the integers 1 to n k are permuted. The simplest example is p(j)=j. Next, a value -s is stored in ~Dq k[i], and the value Dif is decreased by Q k[i]F k[i] in step 524.
This indicates that the quantization value Dq-k[i] is changed by -s by adding ~Dq k[i] to Dq k[i] at adder 64.
The value s is 1 or -1, so the change in the quantization value is one. In other words, de-quantized value obtained in a inverse quantizer in a decoder is changed by -Q k[i], one step size. The value Dif after the update is identical to the absolute value of the difference between cu[k] and Vr[k] calculated with the corrected quantization values Dq k[i] + ~Dq k[i] (i=1,...,n-k). After step 524, the values Dif and zero, and index j and n k are compared in step 526. If Dif < 0 or j>n k then this subroutine is finished. If the condition is not satisfied, the index j is increased by certain amount ~j in step 528, then the process returns to step 523.
Instead of the process shown in Figure 6, an alternative process shown in Figure 7 can be used as the ' '. 27 process of step 506. In the process shown in Figure 7, a step 530, checking whether the quantization value Dq-k[i]
equals zero or not, is added between step 523 and step 524. In this case, step 524 is performed only if the quantization value Dq k[i) is a non-zero value. This allows a reduction in the increase of the number of bits caused by correcting watermark information, because changing a quantization value from zero to non-zero value generally results in a large increase in the number of bits.
Figure 4 is a schematic diagram of an alternative embodiment of a modified MPEG-2 encoder for reducing degradation of a watermark in watermarked data. In this embodiment, the basic concept is the same as the described in connection with Figure 3. The differences lie in the fact that subtraction of the predicted images from the input original images is performed in the DCT
domain not in the spatial domain. For the predicted image outputted from the frame memory 32, DCT is performed in DCT calculator 70, and the results are subtracted from the watermarked DCT coefficients at subtractor 72. The results of subtraction are sent to the quantizer 36 and then the watermark correction is performed in the same manner as shown in Figure 3. The results outputted from subtractor 72 are the same as the results outputted from the DCT calculator 34 in Figure 3 because of the linearity of DCT. Therefore, the results obtained in the ' '. 2 8 processes followed by quantization in quantizer 36 are the same as those in Figure 3. This embodiment results in a reduction in the number of DCT calculations.
The detection procedure to detect a watermark in an image will now be described.
If MPEG video is the input image data format, the following detection process determines whether watermark W is present, where W[1,...,N) is the watermark being tested for.
Decode the Huffman code, but do not compute the inverse DCT's, so that, for each frame (at least, each I-frame), there is an array of 8x8 DCTs.
Next perform the same summation of DCT coefficients that was performed during watermark insertion to obtain the vector V. Compute the correlation coefficient C, between v and the watermark being tested for, W:
vl=v-v w'=w-w _ Finally, convert C into a normalized Fisher Z
statistic:
2o Z= N-31o 1+C
2 ~ 1-C
where N is the length of the watermark.
The Z value indicates whether the watermark is present. A preferred threshold for Z is 4 (i.e. Z >_ 4 means the watermark is present), but other values may be used depending on the desired probabilities of false alarms and missed detec:tions.
Figure 8 is a flow diagram of the detection method for MPEG video input described above. The input MPEG
video is subject to a Huffman decoder and partial parser 80 where the output is a set of DCT for nxn, preferably 8x8, blocks of the video input.
The nxn DCTs are provided to watermark accumulator 82. Accumulator 82 has memory whose length is the watermark length N. DCT coefficients from the Huffman decoder and partial parser 80 are classified. according to a predetermined rule and summed for extracting a watermark as mentioned before, and the results are accumulated in the memory. The extracted watermark is provided to comparator 84 where it is compared with possible watermarks in the image by calculating correlation coefficients between the extracted watermark and the possible watermarks as mentioned before. The possible watermarks are the set of the watermarks that might have been inserted into the video data. The details of the accumulators and comparator are found, for instance, in U.S. Patent 5,915,027 issued June 22, 1999.
The output of cornparator 84 is the li~;elihood (normalized Fisher Z statistic) of the detected watermark being each of the possible watermarks. The most likely watermark is determined and is deemed the watermark in the image, or, if the detector does not exceed a predetermined threshold, then no watermark is present.
Alternatively, if the incoming input data comprises 5 an uncompressed image, an embedded watermark can be detected by applying the method above to DCT coefficients obtained by performing 8x8 DCT for the whole image. In this case, DCT have to be performed for each 8x8 block, but a skillfully designed rule for classifying DCT
10 coefficients into N sets enables us to avoid performing DCT many times. Before explaining the method to reduce DCT calculation, let us define some notations.
Let h m(i,j) (i=1,...,8, j=1,...,8, m=0,...,M-1) be a set of functions that map frequency indices of 8x8 DCT
15 coefficients (i, j) onto the indices k of the element of a watermark, and M indicates the number of functions. So if k=h m(i,j), then a DCT coefficient whose index is (i,j) is classified into the Ic-th set of DCT coefficients .
for calculating the value V[k]. We prepare M different 20 functions h m(i,j) (m=0,...,M-1). Which function is selected for a certain 8x8 block depends on the numbers r and c of the block where r and c indicate the row and column numbers of the block respectively. So the index of the functions, m, is first found according to the values 25 r and c, then the index of the sets, k, is determined by h m(i,j) for each DCT coefficient. Using these functions h m(i,j) in classifying DCT coefficients is assumed in the remaining part of the detailed description.
In this case, we can reduce the number of DCT
calculations in the following manner. First the sum of the blocks whose indices m are the same is computed for each m=0,...,M-1. Let this summed block be denoted. by VB m[i,j] (m=0,...,M-1). Then DCT is perfomred for the M
summed blocks VB m[i,j] (m=0,...,M-1). Fina7.ly, the DCT
coefficients of the summed blocks are classi:Eied into N
sets according to the value h m(i,j), and added together within each set to obtain V[1,...,N]. The obtained results V[1,...,N] are the same as V[1,...,N] obtained with the method mentioned above because DCT .is a linear transform, that is, the sum of DCT blocks equals the result of DCT for the sum of the blocks. If M is much less than the total number of blocks in an image, the number of DCT calculations is dramatic ally reduced. This method thus allows us to extract watermarks with small calculation cost.
Figure 9 shows a flow diagram of the d~=_tection method for uncompressed video input data as described above.
The uncompressed video image data is provided to nxn accumulators, preferably 8x8 accumulators 90. The memory requirement is n2 times the number of the function h m(i,j). For each index m, the blocks with the same index m are summed and the resultant M summed blocks are accumulated in the memory.
The output is the summed signal of each of the nxn blocks. The output is subject to a DCT transform 92. The number of transformations is proportional to the number of the functions h m(i,j). The result is a group of nxn DCTs which is classified into N sets according to the functions h m(i,j) and summed for extracting a watermark as mentioned before. The obtained watermark is provided to watermark accumulator 94 and accumulated. The memory requirement for accumulator 94 is proportional to the , watermark length. The extracted watermark is provided as input to comparator 96. The other inputs to comparator 96 are the possible watermarks that may have been inserted into the input image data. The comparator computes a likelihood (normalized Fisher Z statistic) of each possible watermark having been inserted into the image data. The most likely watermark is determined and is deemed the watermark in the image.
A limitation of block based DCT methods is their sensitivity to spatial shifts of the image. For example, if the image is shifted two pixels to the right, then the DCT coefficients change significantly, so that the watermark cannot be detected. Furthermore, general distortions, such as scaling and rotation, also make the watermark undetectable.
To solve these problems, the above insertion and extraction methods may be modified in two ways. The first possible modification is to insert multiple watermarks designed to survive predefined distortions of the video.
The second modification is to arrange that translations can be compensated for without performing the summation more than once. Optionally, this second modification may be further modified to insert registration patterns, one for each of the multiple watermarks, which can be used by a modified watermark detector to compensate for arbitrary translations of the video.
When detecting watermarks inserted using the above method, it is necessary to divide the image into the same grid of nxn blocks as was used during insertion. If the image has been translated since watermark insertion, then determining the correct grid becomes difficult. In many °
applications, this is a serious problem, since certain, specific transformations can be expected. For example, it can be expected that video on a DVD disk might be modified to fit on a standard television screen by conversion to either "panscan" or "letterbox" mode. In "panscan" mode, the horizontal image resolution is increased, and the image is cropped at a predetermined offset, so that the resulting image will be correct when viewed on a 3x4-aspect-ratio television screen. In "letterbox" mode, the image is scaled vertically, and black is added at the top and bottom, so that the whole image will fit correctly on a 3x4-aspect-ratio screen.
Since these two geometric transformations are more likely than any other, it is reasonable to prepare for them specifically.
The problem of the predetermined scaling or transforming of watermarked video is solved in the present invention by inserting an additional watermark for each of the likely transformations. Each of these watermarks is designed so that, when the image has undergone the corresponding known transformation, the grid of nxn blocks used during insertion will align with a predetermined grid used during detection. Thus, if the image has undergone no transformation, then the detection grid will align with the normal mode watermark, and the normal mode watermark will be detected. If the image has undergone "panscan" transformation, then the same detection grid will align with the "panscan" watermark, and the "panscan" watermark will be detected, and so forth for "letterbox" scan or any other predefined transformation.
The procedure for inserting a watermark that is to be detected after a specific transformation comprises the following steps:
1. Make a copy, IT, of the image being watermarked, I, and apply the transformation to be compensated for the image. For example, IT might be a copy of I that has been transformed into "letterbox" mode by vertical shrinking of I.
2. Create a watermarked version of the transformed image, IT', according to the general watermarking method described above.
3. Let WT = IT'-~IT be the spatial pattern that was added to IT when it was watermarked.
5 4. Perform the inverse transformation on WT to yield the corresponding watermark pattern, W, for the untransformed image. ror example, if the transformation to be compensated for was "letterbox" mode, then W would be obtained by vertically expanding WT.
10 5. Let I' - I + W be the image with a watermark added for the given transformation.
When the transformation is applied to the watermarked image I', the result will be approximately IT', 15 and the watermark will be detected by the same procedure designed to detect a normal watermark.
This process can only be used for a small number of transformations, as each additional watermark causes additional degradation of the image, and reduces the 20 detectability of other watermarks in the image. However, tests have shown that three watermarks --two for transformed images and one for the untransformed image--result in acceptable fidelity and good detectability.
Alternatively, for video, each watermark ca:n be inserted 25 in a time-multiplexed manner.
In cases where the transformations that an image will undergo are not predefined or predetermined, or where there are too many probable transformations to allow for the insertion of a separate watermark for each transformation, the above described method of compensating for transformations is not optimal. Thus, the present invention includes an additional improvement, which compensates for arbitrary translation of the image between the times of watermark insertion and watermark detection.
Arbitrary translation is compensated for in two ways. One way is by translations by even multiples of 8 pixels in the x or y directions when 8x8 blocks are used.
This can be easily compensated for if the following restrictions are imposed on the relationship between an index of function h m(i,j), m, and the row and column numbers of blocks, r and c. We determine the index m according to m = f(r,c) mod M, where f(r,c) is a linear function of r and c, and f(0,1) and f(1,0) are integers.
In addition, let h m(i,j) be a function expressed as h m(i,j) - (h-0(i,j) + axm) mod N, where a is an integer. These restrictions are assumed in the remaining part of the detailed description. In this case, the shift compensation is performed in the basic detection algorithm by computing the correlation of the extracted watermark with all cyclical shifts of the watermark being tested for. This is because the values VB m[i,j] (m=0,...,M-1) obtained from a watermarked image shifted by a multiple of 8 pixels in the horizontal and/or the vertical directions are identical to the values VB m[i,j] (m=0,...,M-1) obtained from the non-shifted image except that the indices m are cyclically shifted. As a result, the extracted value V[1,...,N]
obtained from a watermarked image shifted by a multiple of 8 are identical to the value V[1,...,N] obtained from the non-shifted image except that the elements V[k] are cyclically shifted. For nxn grid format, the same thing is true of a shift of a multiple of n in the horizontal and/or the vertical directions.
For shifts of less than 8 pixels in x and/or y directions, an exhaustive search can be performed of all 64 possibilities and the maximum Z value taken from the set of 64xM - Z values, where M is the number of 8x8 accumulators. In our tests M was chosen to be 64. The factor M is necessary to account for the cyclic shifts that are introduced by shifts of a multiple of 8 pixels.
The exhaustive search requires shifting the M 8x8 accumulator array in the spatial domain and then performing the DCT of each of the M 8x8 blocks. This is performed 64 times, once for each of the possible shifts.
Thus it is necessary to perform 64xM 8x8 DCTs. If this computation is too expensive in terms of time or memory an alternative method can be used, as described below.
The second way, which compensates for translations of non-even multiples of n pixels, uses a pattern (referred to as a "registration pattern") which can be inserted at the time of watermark insertion. By finding the location where the registration signal best matches a predefined signal, a detector can determine how much to shift the data before extracting the watermark. This shifting must be done in the spatial domain, but can be done with accumulators, so conversions of whole images are avoided.
Moreover, the 64xM 8x8 DCTs are unnecessary.
Instead, the correct registration is determined in the spatial domain and then compensated for by shifting the pixels in the accumulator arrays. The M 8x8 accumulators are only then transformed into the DCT domain and the watermark extraction is performed as described above.
The registration pattern is an 8x8 spatial pattern inserted into the image in such a way that the sum of all 8x8 pixel blocks highly correlates with the pattern.
Again, an nxn spatial pattern is used when the video is nxn blocks. A registration pattern can be inserted by using the watermark insertion method described above. The sum of all 8x8 pixel blocks becomes highly correlated with a registration pattern if the DCT coefficients of the sum block and those of the registration pattern are highly correlated to each other. In addition, the DCT
coefficients of the sum block equals the sum of all DCT
blocks because of the linearity of DCT. We can thus insert a registration pattern using a method similar to that inserting a watermark described above, considering the DCT coefficients of a registration pattern as a watermark W and considering the sum of all 8x8 DCT blocks as a value V.
Insertion is performed by converting the registration pattern into the DCT domain, and then using the basic insertion algorithm with different DCT
coefficients Dks and filter coefficients Fks.
Specifically, each AC term of the registration pattern's DCT is considered one element of the watermark, W[k]. The set of DCT terms that are summed together to extract this element, Dk, is simply the set of corresponding terms of all the 8x8 DCTs in the image. All the Fks are set to 1. Using these Dks and Fks, the insertion algorithm inserts a registration pattern along with each watermark. The watermarks are still inserted with the original Dks and Fks.
During detection, registration is performed as follows. Each of a predetermined number of blocks is summed together to form a single nxn, typically 8x8, block.
We have arbitrarily used 64 blocks in our tests. This block contains a registration pattern placed there by the insertion process. To determine the horizontal and vertical translation of the frame, a correlation process is performed in the spatial domain to determine these offsets. Sixty-four correlations are performed for each of the 8 horizontal and 8 vertical motions that are 5 possible. When the 8x8 patch is shifted either horizontally or vertically, a wrap around shift is performed.
We now describe how to determine the shift of the grid. In the following method, we assume the integer 10 values f(0,1) and f(1,0) are relatively prime to M.
First, the blocks for which the same function h m(i,j) is used in classifying DCT coefficients are added together in the spatial domain for generating a summed block VB m[i,j] for each m. The sum of all nxn 15 blocks denoted by AB[i,j], is computed by !1T-1 ~B~l,,j] - ~vB-1111, J~ ~1 - x,...,11, J = 1,...,11.
m=0 Then the correlation coefficient between AB[i,j]
and a registration pattern, denoted by R[i,j], is computed. After calculation of the correlation 20 coefficient, the values AB[i,j] are cyclically shifted by one column in the horizontal direction, and the correlation coefficient between AB[i,j] and R[i,j] is calculated in the same way. The same operations are repeated for each shift. After shifting n times, AB[i,j]
25 becomes identical to AB[i,j] before any shift is done.
Then, the values AB[i,j] are cyclically shifted by one row in the vertical direction, and calculation of a correlation coefficient and shift by one column in the horizontal direction are repeated. In this way, we can calculate correlation coefficients for all n2 possible shifts. At the same time, we search the shift value (offset), denoted by (X, Y), which gives the maximum correlation coefficient.
After the offset (X,Y) has been determined, the M
summed blocks VB m[i,j] are then shifted accordingly in the spatial domain.
Next, we describe the method to compensate for the shift value X in the horizontal direction. To do so, the values VB m[i,j] (m=0,...,M-1) are first copied on an nxnM array VB1[i,j] (i=1,...,n, j=1,...,nM). In copying the values VB m[i,j], the spatial relationships between blocks, that is, which blocks are adjoining a certain block are considered. The function f(r,c) is linear, so f(r,c+1) - f(r,c) + f(0,1).
This indicates that the blocks whose index m equals ml = (m0 + f(0,1)) mod M
are located next to the blocks in which m=m0. So, the values VB ml are copied next to the values VB m0 in the array VB1. For this reason, for each I (I=0,...,M-1), the values VB m'(i,j] (i=1,...,n, j=1,...,n, m'=Ixf(0,1) mod M) are copied in the nxn square region of VB1 where the index of the left-top corner is (1, nI+1). After copying the values for all I, the array VB1 is filled with the values VB m[i,j] because f(0,1) and M are relatively prime. Next, the values in the array VB1 are cyclically shifted by X in the horizontal direction. Then the values in VB1 are returned to VB m in such a way that the value in the nxn region of VB1 where the index of the left-top corner is (1, nI+1) are substituted to VB m'[i,j]
(i=1,...,n, j=1,...,n, m'=Ixf(0,1) mod M) for each I
(I=0,...,M-1). The horizontal offset value X can thus be compensated for without shifting the whole image itself.
Next, the shift value Y in the vertical direction is compensated for. To do so, the values VB m[i,j]
(m=0,...,M-1) are first copied on an nMxn array VB2[i,j]
(i=1,...,nM, j=1,...,n). In copying the values VB m[i,j], the spatial relationship between blocks are considered.
The function f(r,c) is linear, so ' f (r+l, c) - f(r, c) + f (1,0).
This indicates that the blocks whose index m equals ml = (m0 + f(1,0)) mod M
locate under the blocks in which m=m0. So, the values VB ml are copied under the values VB m0 in the array VB2.
For this reason, for each I (I=0,...,M-1), the values VB m'[i,j] (i=1,...,n, j=1,...,n, m'=Ixf(0,1) mod M) are copied on the nxn square region of VB2 where the index of the left-top corner is (nI+1, 1). After copying the values for all I, the array VB2 is filled with the values VB m[i,j] because f(1,0) and M are relatively prime. Next, the values in the array VB2 are cyclically shifted by Y
in the vertical direction. Then the values in VB2 are returned to VB m in such a way that the value in the nxn region of VB2 where the index of the left-top corner is (nI+1, 1) are substituted in VB m'[i,j] (i=1,...n, j=1,...,n, m'=Ixf(0,1) mod M) for each I (I=0,...,M-1).
The vertical offset value Y can thus be compensated without shifting the whole image itself.
The offset of the nxn grid is compensated by the process mentioned above. These processes are performed in the registration process 108 (Figure 10) as will be explained later. A shift of a multiple of n remains even after these processes, but this shift does not affect the watermark detection because the correlation coefficient between a watermark W and an extracted value V is calculated by shifting the watermark W cyclically as described above. After the registration process above is applied, the M blocks VB m (m=0,...,M-1) are transformed back to the DCT domain.
With reference now to Figures 10 and 11, there are shown the basic detection algorithms modified to compensate for translational registration. In the case of MPEG video input (Figure 10), 8x8 DCT blocks obtained from an MPEG video stream are first classified into M
groups according to their indices m of the function h m(i,j), summed within the groups for generating M
summed blocks, and the resultant summed blocks are accumulated in 8x8 accumulators 102. The M summed blocks in accumulators 102 must be converted into the spatial domain by performing an inverse DCT operation in inverse DCT converter 104, and accumulated in accumulators 106.
Finding the offset value of the 8x8 grid and compensating for the offset is executed for the output from 8x8 ' accumulators 106 in registration 108 as described above.
The registration data outputted from registration process 108 is accumulated in accumulators 110 and converted into the DCT domain in DCT converter 112 for watermark extraction by use of accumulators 114, watermark extractor 116 and watermark decoder 118. In watermark extractor 116, the DCT coefficients outputted from accumulator 114 are classified into N sets according to the functions h m(i,j) and summed for extracting a watermark. The obtained watermark is provided to watermark decoder 118, in which the processes executed in comparator 84 in Figure 8 for finding a watermark corresponding to the extracted watermark. The watermark considered to have been inserted is outputted from the watermark decoder 118. In the case of uncompressed input data (Figure 11), the input data is divided into 8x8 blocks and accumulated in accumulators 106 according to the indices of the functions h m(i,j), and registration 108 is performed before conversion into the DCT domain in DCT converter 112. The process continues as described above.
While there has been described and illustrated methods of insertion and detection of watermarks in image 5 data, it will be understood by those skilled in the art that variations and modifications are possible without deviating from the spirit and broad teachings of the present invention which shall be limited solely by the scope of the claims appended hereto.
Claims (37)
1. A method of inserting a watermark signal into data to be watermarked comprising the steps of:
applying a watermark extraction to unwatermarked data for generating extracted values;
computing a difference between a watermark to be inserted and the extracted values; and adding the computed difference throughout the unwatermarked data.
applying a watermark extraction to unwatermarked data for generating extracted values;
computing a difference between a watermark to be inserted and the extracted values; and adding the computed difference throughout the unwatermarked data.
2. A method of inserting a watermarked signal into data to be watermarked as set forth in claim 1, wherein said adding the computed difference comprises distributing the computed difference according to a perceptual model of human visual sensitivity.
3. A method of inserting a watermarked signal into data to be watermarked as set forth in claim 1, wherein said adding the computed difference comprises distributing the computed difference according to how affected the data will be to a predetermined attack so that the attack will not affect the watermarked data.
4. A method of inserting a watermarked signal into data to be watermarked as set forth in claim 1, wherein said adding the computed difference comprises distributing the computed difference according to a stochastic method to reduce the susceptibility to tampering.
5. A method of inserting a watermarked signal into data to be watermarked as set forth in claim 1, wherein said unwatermarked data is compressed video and said compressed video is divided into groups of blocks, where said computing the difference computes a difference between a watermark and the sum of the extracted values from each group of blocks and said adding distributes the difference into the group of blocks.
6. A method of inserting a watermarked signal into data to be watermarked as set forth in claim 5, wherein said groups of blocks are selected according to response to predetermined attacks and signal degradations.
7. A method of detecting a watermark from watermarked data comprising the steps of:
receiving watermarked data where the watermarked data comprises nxn blocks of watermarked data;
summing the nxn blocks of watermarked data to form at least one nxn block of summed watermarked data;
cyclically shifting the block in a spatial. domain to register the block; and extracting the watermark from the at least one nxn block.
receiving watermarked data where the watermarked data comprises nxn blocks of watermarked data;
summing the nxn blocks of watermarked data to form at least one nxn block of summed watermarked data;
cyclically shifting the block in a spatial. domain to register the block; and extracting the watermark from the at least one nxn block.
8. A method of detecting a watermark from watermarked data comprising the steps of:
receiving watermarked data where the watermarked data comprises nxn blocks of watermarked data;
summing the nxn blocks of watermarked data to form at least one nxn block of summed watermarked data; and extracting the watermark from the at least one nxn block by summing predetermined watermarked data from each nxn block to yield each element of the extracted watermark.
receiving watermarked data where the watermarked data comprises nxn blocks of watermarked data;
summing the nxn blocks of watermarked data to form at least one nxn block of summed watermarked data; and extracting the watermark from the at least one nxn block by summing predetermined watermarked data from each nxn block to yield each element of the extracted watermark.
9. A method of detecting a watermark from watermarked data as seat forth in claim 8, wherein said extracting is performed in the spatial domain using at least one nxn blocks.
10. A method of detecting a watermark from watermarked data as set forth in claim 8, wherein said extracting is performed in a frequency domain using at least one nxn blocks.
11. A method of detecting a watermark from watermarked data as set forth in claim 8, wherein each nxn block is filtered before said extracting.
12. A method of detecting a watermark from watermarked data comprising the steps of:
receiving watermarked data where the watermarked data comprises nxn blocks of watermarked data;
summing the nxn blocks of watermarked data to form M
nxn blocks of summed watermarked data, where M is number of functions mapping frequency indices onto indices of elements of watermarks; and extracting the watermark from the at least one nxn block.
receiving watermarked data where the watermarked data comprises nxn blocks of watermarked data;
summing the nxn blocks of watermarked data to form M
nxn blocks of summed watermarked data, where M is number of functions mapping frequency indices onto indices of elements of watermarks; and extracting the watermark from the at least one nxn block.
13. A method of detecting a watermark from watermarked data as set: forth in claim 12, wherein said forming is performed in the spatial domain.
14. A method of detecting a watermark from watermarked data as set: forth in claim 12, wherein said forming is performed in the frequency domain.
15. A method of inserting a watermark signal into video images comprising the steps of:
receiving input video images;
performing discrete cosine transformation (DCT) of said input video images to obtain input video image DCT
values;
adding watermark signals to said input video image DCT values to obtain watermarked DCT values;~
performing an inverse DCT on the watermarked DCT
values for generating watermarked images;
subtracting predicted images stored in memory from said watermarked images for generating residual images;
performing discrete cosine transformation of said residual images to obtain residual image DCT values and quantizing the residual image DCT values to derive residual image quantized DCT values;
inverse quantizing the residual image quantized DCT
values to obtain residual image de-quantized DCT values;
performing discrete cosine transformation of the predicted images to obtain predicted image DCT values;
summing the residual image de-quantized DCT values and the predicted imaged DCT values for generating decoded image DCT coefficients;
calculating correction signals by applying a watermark extraction to the decoded image DCT
coefficients and comparing resultant extracted values with watermark target values;
adding the correction signals to the residual image quantized DCT values for obtaining an output signal;
inverse quantizing the output signal to obtain inverse quantized output data, performing inverse DCT of the inverse quantized output signal and summing the resultant signal with the predicted images for generating a summed signal;
generating predicted images to be stored in said memory by using the summed signal; and variable length encoding the output signal for providing a watermarked MPEG video signal of the input video images.
receiving input video images;
performing discrete cosine transformation (DCT) of said input video images to obtain input video image DCT
values;
adding watermark signals to said input video image DCT values to obtain watermarked DCT values;~
performing an inverse DCT on the watermarked DCT
values for generating watermarked images;
subtracting predicted images stored in memory from said watermarked images for generating residual images;
performing discrete cosine transformation of said residual images to obtain residual image DCT values and quantizing the residual image DCT values to derive residual image quantized DCT values;
inverse quantizing the residual image quantized DCT
values to obtain residual image de-quantized DCT values;
performing discrete cosine transformation of the predicted images to obtain predicted image DCT values;
summing the residual image de-quantized DCT values and the predicted imaged DCT values for generating decoded image DCT coefficients;
calculating correction signals by applying a watermark extraction to the decoded image DCT
coefficients and comparing resultant extracted values with watermark target values;
adding the correction signals to the residual image quantized DCT values for obtaining an output signal;
inverse quantizing the output signal to obtain inverse quantized output data, performing inverse DCT of the inverse quantized output signal and summing the resultant signal with the predicted images for generating a summed signal;
generating predicted images to be stored in said memory by using the summed signal; and variable length encoding the output signal for providing a watermarked MPEG video signal of the input video images.
16. A method of inserting a watermark signal into video images as set forth in claim 15, wherein said calculating step generates a negative value as said correction signal when the extracted value exceeds the corresponding value of said watermark target values, and generates a positive value as said correction signal when the extracted value falls short of the corresponding value of said watermark target values.
17. A method of inserting a watermark signal into video images as set forth in claim 16, wherein said positive value is +1 and said negative value is -1.
18. A method of inserting a watermark signal into video images as set forth in claim 16, wherein said calculating step generates a zero value as the correction signal if the corresponding value of said residual image quantized DCT values is zero.
19. A method of inserting a watermark signal into video images as set forth in claim 18, wherein said positive value is +1 and said negative value is -1.
20. A method of inserting a watermark signal into video images comprising the steps of:
receiving input video images;
performing discrete cosine transformation (DCT) of said input video images to obtain input video image DCT
values;
adding watermark signals to said input video image DCT values to obtain watermarked DCT values;
performing discrete cosine transformation on predicted images stored in memory to obtain predicted image DCT values;
subtracting the predicted image DCT values from the watermarked DCT values to obtain residual image DCT
values;
quantizing the residual image DCT values to obtain residual image quantized DCT values;
inverse quantizing the residual image quantized DCT
values to obtain residual image de-quantized DCT values and summing the residual image de-quantized DCT values with the predicted image DCT values, for generating decoded image DCT coefficients;
calculating correction signals by applying a watermark extraction to the decoded image DCT
coefficients and comparing resultant extracted values with watermark target values;
summing the correction signals with the residual image quantized DCT values to obtain an output signal;
inverse quantizing the output signal to obtain inverse quantized output signal, performing inverse DCT
of the inverse quantized output signal and summing the resultant signal with the predicted images to provide locally-decoded images;
generating predicted images to be stored in said memory using the locally-decoded images; and valuable length encoding the output signal to provide a watermarked MPEG video,signal of the input video images.
receiving input video images;
performing discrete cosine transformation (DCT) of said input video images to obtain input video image DCT
values;
adding watermark signals to said input video image DCT values to obtain watermarked DCT values;
performing discrete cosine transformation on predicted images stored in memory to obtain predicted image DCT values;
subtracting the predicted image DCT values from the watermarked DCT values to obtain residual image DCT
values;
quantizing the residual image DCT values to obtain residual image quantized DCT values;
inverse quantizing the residual image quantized DCT
values to obtain residual image de-quantized DCT values and summing the residual image de-quantized DCT values with the predicted image DCT values, for generating decoded image DCT coefficients;
calculating correction signals by applying a watermark extraction to the decoded image DCT
coefficients and comparing resultant extracted values with watermark target values;
summing the correction signals with the residual image quantized DCT values to obtain an output signal;
inverse quantizing the output signal to obtain inverse quantized output signal, performing inverse DCT
of the inverse quantized output signal and summing the resultant signal with the predicted images to provide locally-decoded images;
generating predicted images to be stored in said memory using the locally-decoded images; and valuable length encoding the output signal to provide a watermarked MPEG video,signal of the input video images.
21. A method of inserting a watermark signal into video images as set forth in claim 20, wherein said calculating step generates a negative value as said correction signal when the extracted value exceeds the corresponding value of said watermark target values, and generates a positive value as said correction signal when the extracted value falls short of the corresponding value of said watermark target values.
22. A method of inserting a watermark signal into video images set forth in claim 21, wherein said positive value is + 1 and said negative value is -1.
23. A method of inserting a watermark signal into video images as set forth in claim 21, wherein said calculating step generates a zero value as the correction signal if the corresponding value of said residual image quantized DCT values is zero.
24. A method of inserting a watermark signal into video images as set forth in claim 23, wherein said positive value is +1 and said negative value is -1.
25. A method of inserting an extractable watermark signal into image data so that the watermark signal can be extracted after the watermarked data is subject to distortion comprising the steps of:
providing image data to be watermarked;
copying the image data;
applying a predefined distortion to the copy of image data;
inserting a watermark signal into the distorted copy of image data;
subtracting the copy of distorted image data from the watermarked copy of distorted image data to generate a watermark pattern;
performing inverse distortion on the watermark pattern to generate the watermarked pattern for nondistorted image data; and combining said image data to be watermarked and said watermark pattern for nondistorted image data to obtain watermarked data.
providing image data to be watermarked;
copying the image data;
applying a predefined distortion to the copy of image data;
inserting a watermark signal into the distorted copy of image data;
subtracting the copy of distorted image data from the watermarked copy of distorted image data to generate a watermark pattern;
performing inverse distortion on the watermark pattern to generate the watermarked pattern for nondistorted image data; and combining said image data to be watermarked and said watermark pattern for nondistorted image data to obtain watermarked data.
26. A method of inserting a watermark signal into image data as set forth in claim 25, wherein said predetermined distortion is panscan mode image data.
27. A method of inserting a watermark signal into image data as set forth in claim 25, wherein said predetermined distortion is letterbox mode image data.
28. A method of inserting a watermark signal into image data as set forth in claim 25, wherein said combining includes combining with perceptual modeling.
29. A method of inserting an extractable watermark signal into image data so that the watermark signal can be extracted after the watermarked data is subject to distortion comprising the steps of:
receiving image data in form of nxn blocks; and inserting an nxn registration pattern into the image data.
receiving image data in form of nxn blocks; and inserting an nxn registration pattern into the image data.
30. A method of inserting an extractable watermark signal into image data as set forth in claim 29, wherein said inserting is performed by converting the registration pattern into the DCT domain and then inserting the DCT values of the registration pattern into the image data.
31. A method of detecting a watermark from watermarked image data that has been subjected to a two dimensional translation distortion comprising the steps of:
partitioning received distorted watermarked image data into a set of nxn blocks;
summing together watermarked image data in a predetermined number of nxn blocks;
if summed nxn blocks are not in spatial domain, transforming summed nxn blocks into spatial domain;
determining two-dimensional translation distortion of the summed nxn blocks;
shifting the summed nxn blocks according to the determined distortion;
transforming the shifted nxn blocks into transform domain;
extracting a possible watermark signal from the transformed nxn blocks;
correlating the possible watermark with a known watermark;
repeating said correlating for cyclic shifts of the possible watermark; and determining the presence of a watermark based on said correlating.
partitioning received distorted watermarked image data into a set of nxn blocks;
summing together watermarked image data in a predetermined number of nxn blocks;
if summed nxn blocks are not in spatial domain, transforming summed nxn blocks into spatial domain;
determining two-dimensional translation distortion of the summed nxn blocks;
shifting the summed nxn blocks according to the determined distortion;
transforming the shifted nxn blocks into transform domain;
extracting a possible watermark signal from the transformed nxn blocks;
correlating the possible watermark with a known watermark;
repeating said correlating for cyclic shifts of the possible watermark; and determining the presence of a watermark based on said correlating.
32. A method of detecting a watermark from watermarked image data as set forth in claim 31, wherein said nxn block is a 8x8 block.
33. A method of detecting a watermark from watermarked image data as set forth in claim 31, where the two dimensional translation is a geometric shift in at least one of the horizontal or vertical direction.
34. A method of detecting a watermark from watermarked image data as set forth in claim 31, wherein said correlating is performed for n2 possible shifts.
35. A method of detecting a watermark from watermarked image data that has been subjected to a two dimensional translation distortion comprising the steps of:
(a) partitioning received distorted watermarked image data in nxn blocks;
(b) summing together watermarked image data in a predetermined number of nxn blocks;
(c) if summed nxn blocks are not in spatial domain, transforming summed nxn blocks into spatial domain;
(d) shifting the summed nxn blocks by one of n2 possible shifts;
(e) transforming the shifted nxn blocks into transform domain;
(f) extracting a possible watermark signal from the transformed nxn blocks;
(g) correlating the possible watermark with a known watermark;
(h) repeating said correlating for cyclic shifts of the possible watermark;
(i) repeating steps (d)-(h) for each of the n2 possible shifts; and (j) determining the presence of a watermark based on the results of step (i).
(a) partitioning received distorted watermarked image data in nxn blocks;
(b) summing together watermarked image data in a predetermined number of nxn blocks;
(c) if summed nxn blocks are not in spatial domain, transforming summed nxn blocks into spatial domain;
(d) shifting the summed nxn blocks by one of n2 possible shifts;
(e) transforming the shifted nxn blocks into transform domain;
(f) extracting a possible watermark signal from the transformed nxn blocks;
(g) correlating the possible watermark with a known watermark;
(h) repeating said correlating for cyclic shifts of the possible watermark;
(i) repeating steps (d)-(h) for each of the n2 possible shifts; and (j) determining the presence of a watermark based on the results of step (i).
36. A method of detecting a watermark from watermarked image data as set forth in claim 35, wherein said nxn block is a 8x8 block.
37. A method of detecting a watermark from watermarked image data as set forth in claim 35, wherein said correlating is performed for n2 possible shifts.
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US9053298P | 1998-06-24 | 1998-06-24 | |
US60/090,532 | 1998-06-24 | ||
US09/118,467 US6154571A (en) | 1998-06-24 | 1998-07-17 | Robust digital watermarking |
US09/118,467 | 1998-07-17 |
Publications (2)
Publication Number | Publication Date |
---|---|
CA2276378A1 CA2276378A1 (en) | 1999-12-24 |
CA2276378C true CA2276378C (en) | 2003-02-04 |
Family
ID=26782379
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CA002276378A Expired - Fee Related CA2276378C (en) | 1998-06-24 | 1999-06-23 | Robust digital watermarking |
Country Status (4)
Country | Link |
---|---|
US (4) | US6154571A (en) |
EP (4) | EP2306726B1 (en) |
JP (5) | JP3687725B2 (en) |
CA (1) | CA2276378C (en) |
Families Citing this family (248)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6345104B1 (en) * | 1994-03-17 | 2002-02-05 | Digimarc Corporation | Digital watermarks and methods for security documents |
US7171016B1 (en) * | 1993-11-18 | 2007-01-30 | Digimarc Corporation | Method for monitoring internet dissemination of image, video and/or audio files |
US6424725B1 (en) * | 1996-05-16 | 2002-07-23 | Digimarc Corporation | Determining transformations of media signals with embedded code signals |
US20020136429A1 (en) * | 1994-03-17 | 2002-09-26 | John Stach | Data hiding through arrangement of objects |
US7362775B1 (en) | 1996-07-02 | 2008-04-22 | Wistaria Trading, Inc. | Exchange mechanisms for digital information packages with bandwidth securitization, multichannel digital watermarks, and key management |
US7805500B2 (en) | 1995-05-08 | 2010-09-28 | Digimarc Corporation | Network linking methods and apparatus |
US6728390B2 (en) * | 1995-05-08 | 2004-04-27 | Digimarc Corporation | Methods and systems using multiple watermarks |
US6763123B2 (en) | 1995-05-08 | 2004-07-13 | Digimarc Corporation | Detection of out-of-phase low visibility watermarks |
US6738495B2 (en) * | 1995-05-08 | 2004-05-18 | Digimarc Corporation | Watermarking enhanced to withstand anticipated corruptions |
US20090097695A9 (en) * | 1995-05-08 | 2009-04-16 | Rhoads Geoffrey B | Personal document authentication system using watermarking |
US7054462B2 (en) * | 1995-05-08 | 2006-05-30 | Digimarc Corporation | Inferring object status based on detected watermark data |
US5613004A (en) | 1995-06-07 | 1997-03-18 | The Dice Company | Steganographic method and device |
US7289643B2 (en) | 2000-12-21 | 2007-10-30 | Digimarc Corporation | Method, apparatus and programs for generating and utilizing content signatures |
US6965682B1 (en) | 1999-05-19 | 2005-11-15 | Digimarc Corp | Data transmission by watermark proxy |
US6505160B1 (en) | 1995-07-27 | 2003-01-07 | Digimarc Corporation | Connected audio and other media objects |
US7711564B2 (en) | 1995-07-27 | 2010-05-04 | Digimarc Corporation | Connected audio and other media objects |
US7006661B2 (en) * | 1995-07-27 | 2006-02-28 | Digimarc Corp | Digital watermarking systems and methods |
US7664263B2 (en) | 1998-03-24 | 2010-02-16 | Moskowitz Scott A | Method for combining transfer functions with predetermined key creation |
US6205249B1 (en) | 1998-04-02 | 2001-03-20 | Scott A. Moskowitz | Multiple transform utilization and applications for secure digital watermarking |
US20030056103A1 (en) | 2000-12-18 | 2003-03-20 | Levy Kenneth L. | Audio/video commerce application architectural framework |
US7095874B2 (en) | 1996-07-02 | 2006-08-22 | Wistaria Trading, Inc. | Optimization methods for the insertion, protection, and detection of digital watermarks in digitized data |
US7159116B2 (en) | 1999-12-07 | 2007-01-02 | Blue Spike, Inc. | Systems, methods and devices for trusted transactions |
US5889868A (en) | 1996-07-02 | 1999-03-30 | The Dice Company | Optimization methods for the insertion, protection, and detection of digital watermarks in digitized data |
US7457962B2 (en) | 1996-07-02 | 2008-11-25 | Wistaria Trading, Inc | Optimization methods for the insertion, protection, and detection of digital watermarks in digitized data |
US7346472B1 (en) | 2000-09-07 | 2008-03-18 | Blue Spike, Inc. | Method and device for monitoring and analyzing signals |
US7177429B2 (en) | 2000-12-07 | 2007-02-13 | Blue Spike, Inc. | System and methods for permitting open access to data objects and for securing data within the data objects |
US7123718B1 (en) * | 1999-03-24 | 2006-10-17 | Blue Spike, Inc. | Utilizing data reduction in stegnographic and cryptographic systems |
US7730317B2 (en) | 1996-12-20 | 2010-06-01 | Wistaria Trading, Inc. | Linear predictive coding implementation of digital watermarks |
JPH1132200A (en) * | 1997-07-09 | 1999-02-02 | Matsushita Electric Ind Co Ltd | Watermark data insertion method and watermark data detection method |
US6108434A (en) * | 1997-09-12 | 2000-08-22 | Signafy, Inc. | Counteracting geometric distortions for DCT based watermarking |
US6971011B1 (en) * | 1998-03-04 | 2005-11-29 | Koninklijke Philips Electronics N.V. | Watermark detection |
US7602940B2 (en) | 1998-04-16 | 2009-10-13 | Digimarc Corporation | Steganographic data hiding using a device clock |
US7644282B2 (en) | 1998-05-28 | 2010-01-05 | Verance Corporation | Pre-processed information embedding system |
TW395133B (en) * | 1998-07-15 | 2000-06-21 | Koninkl Philips Electronics Nv | Detection of a watermark in a compressed video signal |
JP3266569B2 (en) * | 1998-07-29 | 2002-03-18 | 日本電気株式会社 | Image attribute change system using digital watermark data |
US6704431B1 (en) * | 1998-09-04 | 2004-03-09 | Nippon Telegraph And Telephone Corporation | Method and apparatus for digital watermarking |
JP2000092461A (en) * | 1998-09-10 | 2000-03-31 | Sony Corp | Additional information superimposing method and video signal output device |
US7532740B2 (en) | 1998-09-25 | 2009-05-12 | Digimarc Corporation | Method and apparatus for embedding auxiliary information within original data |
US7197156B1 (en) | 1998-09-25 | 2007-03-27 | Digimarc Corporation | Method and apparatus for embedding auxiliary information within original data |
US7373513B2 (en) | 1998-09-25 | 2008-05-13 | Digimarc Corporation | Transmarking of multimedia signals |
US6778678B1 (en) * | 1998-10-02 | 2004-08-17 | Lucent Technologies, Inc. | High-capacity digital image watermarking based on waveform modulation of image components |
JP4001717B2 (en) * | 1998-10-29 | 2007-10-31 | コーニンクレッカ フィリップス エレクトロニクス エヌ ヴィ | Watermark detection method and apparatus |
CA2288366A1 (en) * | 1998-11-05 | 2000-05-05 | Akira Ogino | Additional information transmission method, additional information transmission system, information signal output apparatus, information signal processing apparatus, information signal recording apparatus and information signal recording medium |
ATE289435T1 (en) * | 1998-11-19 | 2005-03-15 | Digimarc Corp | PRINTING AND VERIFICATION OF SELF-CHECKING SECURITY DOCUMENTS |
FR2787604B1 (en) * | 1998-11-30 | 2001-08-31 | Canon Kk | METHOD FOR INSERTING A SECRET MARK AND AUTHENTICATING A DIGITAL SIGNAL |
JP3397157B2 (en) | 1999-01-13 | 2003-04-14 | 日本電気株式会社 | Digital watermark insertion system |
JP3942759B2 (en) * | 1999-01-25 | 2007-07-11 | 興和株式会社 | Digital watermark embedding method, decoding method and apparatus |
US6792128B1 (en) * | 1999-02-16 | 2004-09-14 | Canon Kabushiki Kaisha | Method of inserting and extracting a digital signature |
US6546113B1 (en) * | 1999-03-02 | 2003-04-08 | Leitch Technology International Inc. | Method and apparatus for video watermarking |
US6522766B1 (en) * | 1999-03-15 | 2003-02-18 | Seiko Epson Corporation | Watermarking with random zero-mean patches for copyright protection |
US6556688B1 (en) * | 1999-03-15 | 2003-04-29 | Seiko Epson Corporation | Watermarking with random zero-mean patches for printer tracking |
WO2000056058A1 (en) * | 1999-03-18 | 2000-09-21 | British Broadcasting Corporation | Watermarking |
US6442284B1 (en) * | 1999-03-19 | 2002-08-27 | Digimarc Corporation | Watermark detection utilizing regions with higher probability of success |
JP4470322B2 (en) * | 1999-03-19 | 2010-06-02 | ソニー株式会社 | Additional information embedding method and apparatus, additional information demodulation method and demodulating apparatus |
US7013021B2 (en) * | 1999-03-19 | 2006-03-14 | Digimarc Corporation | Watermark detection utilizing regions with higher probability of success |
US7664264B2 (en) | 1999-03-24 | 2010-02-16 | Blue Spike, Inc. | Utilizing data reduction in steganographic and cryptographic systems |
US7031491B1 (en) * | 1999-04-09 | 2006-04-18 | Canon Kabushiki Kaisha | Method for determining a partition in order to insert a watermark, and associated insertion and decoding methods |
JP3695217B2 (en) * | 1999-04-30 | 2005-09-14 | オムロン株式会社 | Image processing apparatus and image input apparatus |
US8874244B2 (en) | 1999-05-19 | 2014-10-28 | Digimarc Corporation | Methods and systems employing digital content |
US7185201B2 (en) | 1999-05-19 | 2007-02-27 | Digimarc Corporation | Content identifiers triggering corresponding responses |
JP2001045448A (en) * | 1999-07-30 | 2001-02-16 | Nec Corp | Video data synchronization system for digital tv broadcast |
US7475246B1 (en) | 1999-08-04 | 2009-01-06 | Blue Spike, Inc. | Secure personal content server |
US6654479B1 (en) * | 1999-08-19 | 2003-11-25 | Academia Sinica | Cocktail watermarking on images |
US6834344B1 (en) * | 1999-09-17 | 2004-12-21 | International Business Machines Corporation | Semi-fragile watermarks |
WO2001031630A1 (en) * | 1999-10-29 | 2001-05-03 | Sony Corporation | Method and apparatus for data processing, and medium storing program |
US7272718B1 (en) * | 1999-10-29 | 2007-09-18 | Sony Corporation | Device, method and storage medium for superimposing first and second watermarking information on an audio signal based on psychological auditory sense analysis |
US6959098B1 (en) * | 1999-11-30 | 2005-10-25 | Digimarc Corporation | Method and system for determining image transformation |
US6665418B1 (en) * | 1999-12-01 | 2003-12-16 | Eastman Kodak Company | Linear projective data embedding |
US6606393B1 (en) * | 1999-12-02 | 2003-08-12 | Verizon Laboratories Inc. | Message authentication code using image histograms |
JP2001189844A (en) * | 1999-12-28 | 2001-07-10 | Nec Corp | Information inserting and detecting system |
US6915481B1 (en) * | 2000-01-11 | 2005-07-05 | Cognicity, Inc. | Transactional watermarking |
JP3567975B2 (en) | 2000-01-24 | 2004-09-22 | 日本電気株式会社 | Digital watermark detection / insertion device |
US7142689B2 (en) * | 2000-01-31 | 2006-11-28 | Canon Kabushiki Kaisha | Image processing apparatus for determining specific images |
US6737957B1 (en) * | 2000-02-16 | 2004-05-18 | Verance Corporation | Remote control signaling using audio watermarks |
EP1186161A1 (en) * | 2000-02-29 | 2002-03-13 | Koninklijke Philips Electronics N.V. | Embedding and detecting a watermark in an information signal |
US7127744B2 (en) * | 2000-03-10 | 2006-10-24 | Digimarc Corporation | Method and apparatus to protect media existing in an insecure format |
US6804356B1 (en) * | 2000-03-20 | 2004-10-12 | Koninklijke Philips Electronics N.V. | Hierarchical authentication system for images and video |
JP2001275115A (en) * | 2000-03-23 | 2001-10-05 | Nec Corp | Electronic watermark data insertion device and detector |
US7046808B1 (en) * | 2000-03-24 | 2006-05-16 | Verance Corporation | Method and apparatus for detecting processing stages applied to a signal |
US8091025B2 (en) | 2000-03-24 | 2012-01-03 | Digimarc Corporation | Systems and methods for processing content objects |
WO2001073567A1 (en) * | 2000-03-28 | 2001-10-04 | Oak Technology, Inc. | Secure compact disc technology |
JP2001285607A (en) * | 2000-03-29 | 2001-10-12 | Nec Corp | Electronic watermark insertion device, electronic watermark detector, and electronic watermark insertion method and electronic watermark detection method used therefor |
US6707926B1 (en) * | 2000-03-31 | 2004-03-16 | Intel Corporation | Template for watermark decoder synchronization |
US6891959B2 (en) | 2000-04-19 | 2005-05-10 | Digimarc Corporation | Hiding information out-of-phase in color channels |
US8027509B2 (en) | 2000-04-19 | 2011-09-27 | Digimarc Corporation | Digital watermarking in data representing color channels |
US7738673B2 (en) | 2000-04-19 | 2010-06-15 | Digimarc Corporation | Low visible digital watermarks |
US7305104B2 (en) | 2000-04-21 | 2007-12-04 | Digimarc Corporation | Authentication of identification documents using digital watermarks |
US8121843B2 (en) | 2000-05-02 | 2012-02-21 | Digimarc Corporation | Fingerprint methods and systems for media signals |
JP2001319416A (en) * | 2000-05-12 | 2001-11-16 | Pioneer Electronic Corp | Information detecting device |
US7508944B1 (en) * | 2000-06-02 | 2009-03-24 | Digimarc Corporation | Using classification techniques in digital watermarking |
KR100611094B1 (en) | 2000-06-15 | 2006-08-09 | 주식회사 케이티 | Apparatus and method for inserting/detecting watermark based stochastic model |
US6633654B2 (en) * | 2000-06-19 | 2003-10-14 | Digimarc Corporation | Perceptual modeling of media signals based on local contrast and directional edges |
US6594373B1 (en) * | 2000-07-19 | 2003-07-15 | Digimarc Corporation | Multi-carrier watermarks using carrier signals modulated with auxiliary messages |
US6512836B1 (en) * | 2000-07-28 | 2003-01-28 | Verizon Laboratories Inc. | Systems and methods for etching digital watermarks |
US7127615B2 (en) | 2000-09-20 | 2006-10-24 | Blue Spike, Inc. | Security based on subliminal and supraliminal channels for data objects |
US6611608B1 (en) * | 2000-10-18 | 2003-08-26 | Matsushita Electric Industrial Co., Ltd. | Human visual model for data hiding |
EP1220152A3 (en) * | 2000-12-07 | 2004-11-24 | Sony United Kingdom Limited | Embedding data in material |
EP1215880A3 (en) | 2000-12-07 | 2003-08-13 | Sony United Kingdom Limited | Embedding data in material |
GB2369949B (en) * | 2000-12-07 | 2005-03-16 | Sony Uk Ltd | Embedding data in material |
JP3682222B2 (en) * | 2000-12-11 | 2005-08-10 | 株式会社東芝 | Digital watermark detection apparatus, digital watermark detection method, and recording medium |
GB2370437A (en) * | 2000-12-20 | 2002-06-26 | Motorola Inc | Method for watermarking a digital image |
US7095870B2 (en) * | 2000-12-21 | 2006-08-22 | Hitachi, Ltd. | Electronic watermark embedding apparatus and method and a format conversion device having a watermark embedding function |
KR100355033B1 (en) * | 2000-12-30 | 2002-10-19 | 주식회사 실트로닉 테크놀로지 | Apparatus and Method for Watermark Embedding and Detection using the Linear Prediction Analysis |
US7058215B2 (en) * | 2001-01-18 | 2006-06-06 | Canon Kabushiki Kaisha | Data processing system, maintenance apparatus and image processing apparatus |
MXPA02009217A (en) * | 2001-01-23 | 2003-05-23 | Koninkl Philips Electronics Nv | Watermarking a compressed information signal. |
WO2002065753A1 (en) * | 2001-02-13 | 2002-08-22 | Koninklijke Philips Electronics N.V. | Method and arrangement for detecting a watermark in an information signal |
US20030156735A1 (en) * | 2001-02-13 | 2003-08-21 | Maes Maurice Jerome Justin Jean-Baptiste | Method and arrangement for detecting multiple watermarks in an information signal |
US7043051B2 (en) * | 2001-02-21 | 2006-05-09 | Lg Electronics Inc. | Proprietary watermark system for secure digital media and content distribution |
US7602936B2 (en) * | 2001-03-08 | 2009-10-13 | Sony Corporation | Method to make wavelet watermarks resistant to affine transformations |
US7058201B2 (en) * | 2001-03-28 | 2006-06-06 | Lg Electronics Inc. | Method of embedding watermark into digital image |
US6721362B2 (en) | 2001-03-30 | 2004-04-13 | Redrock Semiconductor, Ltd. | Constrained discrete-cosine-transform coefficients for better error detection in a corrupted MPEG-4 bitstreams |
KR100374665B1 (en) * | 2001-04-13 | 2003-03-04 | 주식회사 마크애니 | Method of inserting/detecting digital watermarks and apparatus for using thereof |
JP2003009102A (en) * | 2001-04-18 | 2003-01-10 | Victor Co Of Japan Ltd | Coding method of object data, transmitting method of object data, decoding method of object data, object data coding equipment, object data producing equipment, object data decoding equipment, program for coding object data, program for decoding object data and object data recording medium |
KR100423159B1 (en) * | 2001-04-20 | 2004-03-18 | 주식회사 마크애니 | Method of inserting/detecting digital watermarks and apparatus for using thereof |
KR100378222B1 (en) * | 2001-04-21 | 2003-03-29 | 주식회사 마크애니 | Method of inserting/detecting digital watermarks and apparatus for using thereof |
US7046819B2 (en) | 2001-04-25 | 2006-05-16 | Digimarc Corporation | Encoded reference signal for digital watermarks |
JP2002330279A (en) * | 2001-05-07 | 2002-11-15 | Techno Mathematical Co Ltd | Method for embedding data in image and method for extracting the data |
KR100493284B1 (en) * | 2001-05-11 | 2005-06-03 | 엘지전자 주식회사 | Copy protection method and system for digital media |
JP3565182B2 (en) | 2001-05-11 | 2004-09-15 | 日本電気株式会社 | Method and method for preventing input of variable length code from being interrupted |
US7496197B2 (en) * | 2001-06-14 | 2009-02-24 | Portauthority Technologies Inc. | Method and system for robust embedding of watermarks and steganograms in digital video content |
US20030016842A1 (en) * | 2001-07-23 | 2003-01-23 | Patton Charles M. | Digital content with information of latent value to purchaser and method for making the same |
US7203335B2 (en) * | 2001-08-07 | 2007-04-10 | Gabriel Fielding | System and method for extracting a watermark signal in a digital image sequence |
JP3675745B2 (en) * | 2001-08-21 | 2005-07-27 | ソニー株式会社 | Additional data embedding device, additional data embedding method, and additional data embedding program |
KR20040029029A (en) * | 2001-08-24 | 2004-04-03 | 코닌클리케 필립스 일렉트로닉스 엔.브이. | Adding fields of a video frame |
US7095872B2 (en) * | 2001-08-28 | 2006-08-22 | University Of North Carolina At Charlotte | Automated digital watermarking methods using neural networks |
EP1433175A1 (en) * | 2001-09-05 | 2004-06-30 | Koninklijke Philips Electronics N.V. | A robust watermark for dsd signals |
US7062045B2 (en) * | 2001-09-06 | 2006-06-13 | Clwt, Llc | Media protection system and method |
US20030046568A1 (en) * | 2001-09-06 | 2003-03-06 | Riddick Christopher J. | Media protection system and method and hardware decryption module used therein |
JP3774651B2 (en) * | 2001-09-25 | 2006-05-17 | キヤノン株式会社 | Image processing apparatus and method, computer program, and recording medium |
US6724914B2 (en) * | 2001-10-16 | 2004-04-20 | Digimarc Corporation | Progressive watermark decoding on a distributed computing platform |
US6975745B2 (en) * | 2001-10-25 | 2005-12-13 | Digimarc Corporation | Synchronizing watermark detectors in geometrically distorted signals |
US20030088773A1 (en) * | 2001-11-07 | 2003-05-08 | Koninklijke Philips Electronics N. V. | Method of and apparatus for preventing illicit copying of digital content |
US7302575B2 (en) * | 2001-11-07 | 2007-11-27 | Koninklijke Philips Electronics N.V. | Apparatus for and method of preventing illicit copying of digital content |
US6563937B1 (en) * | 2001-11-28 | 2003-05-13 | Sony Corporation | Method and apparatus to detect watermark that are resistant to arbitrary deformations |
US7433489B2 (en) | 2001-11-28 | 2008-10-07 | Sony Electronics Inc. | Method to ensure temporal synchronization and reduce complexity in the detection of temporal watermarks |
GB2383218A (en) * | 2001-12-13 | 2003-06-18 | Sony Uk Ltd | Watermarking using cyclic shifting of code words |
GB2383148A (en) * | 2001-12-13 | 2003-06-18 | Sony Uk Ltd | Watermarking |
GB2383221A (en) * | 2001-12-13 | 2003-06-18 | Sony Uk Ltd | Method of identifying a codeword used as a watermark |
GB2383220B (en) * | 2001-12-13 | 2005-11-30 | Sony Uk Ltd | Data processing apparatus and method |
JP4491237B2 (en) | 2001-12-21 | 2010-06-30 | コーニンクレッカ フィリップス エレクトロニクス エヌ ヴィ | Improving the integrity of watermarks that use robust features |
KR100454668B1 (en) * | 2001-12-27 | 2004-11-05 | 한국전자통신연구원 | System for embedding and detecting watermark including template |
US20030131350A1 (en) | 2002-01-08 | 2003-07-10 | Peiffer John C. | Method and apparatus for identifying a digital audio signal |
US7321667B2 (en) * | 2002-01-18 | 2008-01-22 | Digimarc Corporation | Data hiding through arrangement of objects |
US7231061B2 (en) * | 2002-01-22 | 2007-06-12 | Digimarc Corporation | Adaptive prediction filtering for digital watermarking |
CN100362586C (en) * | 2002-02-19 | 2008-01-16 | 皇家飞利浦电子股份有限公司 | Collusion-robust watermarking |
US7424128B2 (en) | 2002-04-10 | 2008-09-09 | Pioneer Corporation | Electronic watermark embedding apparatus, electronic watermark embedding method, record medium having electronic watermark and manufacturing method of record medium |
US7287275B2 (en) | 2002-04-17 | 2007-10-23 | Moskowitz Scott A | Methods, systems and devices for packet watermarking and efficient provisioning of bandwidth |
CN1669046B (en) | 2002-05-14 | 2012-03-14 | 施赖纳集团两合公司 | Visible authentication patterns for printed document |
US6782116B1 (en) * | 2002-11-04 | 2004-08-24 | Mediasec Technologies, Gmbh | Apparatus and methods for improving detection of watermarks in content that has undergone a lossy transformation |
JP3958128B2 (en) * | 2002-06-28 | 2007-08-15 | キヤノン株式会社 | Image processing apparatus and method, computer program, and computer-readable storage medium |
JP3919612B2 (en) * | 2002-06-28 | 2007-05-30 | キヤノン株式会社 | Image processing apparatus and method, computer program, and computer-readable storage medium |
JP4054619B2 (en) * | 2002-06-28 | 2008-02-27 | キヤノン株式会社 | Image processing apparatus and method, computer program, and computer-readable storage medium |
JP3919613B2 (en) * | 2002-06-28 | 2007-05-30 | キヤノン株式会社 | Image processing apparatus and method, computer program, and computer-readable storage medium |
WO2004006559A2 (en) * | 2002-07-09 | 2004-01-15 | Kaleidescape, Inc. | Secure presentation of encrypted digital content |
US7036024B2 (en) * | 2002-07-09 | 2006-04-25 | Kaleidescape, Inc. | Detecting collusion among multiple recipients of fingerprinted information |
US20040091111A1 (en) * | 2002-07-16 | 2004-05-13 | Levy Kenneth L. | Digital watermarking and fingerprinting applications |
US7133534B2 (en) * | 2002-09-03 | 2006-11-07 | Koninklijke Philips Electronics N.V. | Copy protection via redundant watermark encoding |
ES2507642T3 (en) | 2002-10-15 | 2014-10-15 | Verance Corporation | Media supervision, management and information system |
CA2503340A1 (en) | 2002-10-23 | 2004-05-06 | Arun Ramaswamy | Digital data insertion apparatus and methods for use with compressed audio/video data |
JP3922369B2 (en) * | 2003-01-21 | 2007-05-30 | 日本ビクター株式会社 | Embedded information recording apparatus and reproducing apparatus, recording program, and reproducing program |
CN1802665A (en) * | 2003-05-09 | 2006-07-12 | 阿卫德技术公司 | Watermarking |
US7460684B2 (en) * | 2003-06-13 | 2008-12-02 | Nielsen Media Research, Inc. | Method and apparatus for embedding watermarks |
US7574063B2 (en) * | 2003-07-23 | 2009-08-11 | Canon Kabushiki Kaisha | Image coding method and apparatus |
US7239748B2 (en) * | 2003-08-12 | 2007-07-03 | International Business Machines Corporation | System and method for segmenting an electronic image |
US8301893B2 (en) | 2003-08-13 | 2012-10-30 | Digimarc Corporation | Detecting media areas likely of hosting watermarks |
US20060239501A1 (en) * | 2005-04-26 | 2006-10-26 | Verance Corporation | Security enhancements of digital watermarks for multi-media content |
US7616776B2 (en) | 2005-04-26 | 2009-11-10 | Verance Corproation | Methods and apparatus for enhancing the robustness of watermark extraction from digital host content |
US9055239B2 (en) | 2003-10-08 | 2015-06-09 | Verance Corporation | Signal continuity assessment using embedded watermarks |
US7369677B2 (en) | 2005-04-26 | 2008-05-06 | Verance Corporation | System reactions to the detection of embedded watermarks in a digital host content |
JP4208135B2 (en) * | 2003-11-07 | 2009-01-14 | Kddi株式会社 | Drift compensation method |
WO2005099385A2 (en) | 2004-04-07 | 2005-10-27 | Nielsen Media Research, Inc. | Data insertion apparatus and methods for use with compressed audio/video data |
US7668334B2 (en) | 2004-07-02 | 2010-02-23 | Digimarc Corp | Conditioning imagery to better receive steganographic encoding |
CN1993700B (en) | 2004-07-02 | 2012-03-14 | 尼尔逊媒介研究股份有限公司 | Methods and apparatus for mixing compressed digital bit streams |
SG120173A1 (en) * | 2004-08-17 | 2006-03-28 | Sony Corp | Methods and apparatus for watermarking digital data |
JP4624359B2 (en) * | 2004-09-06 | 2011-02-02 | 三菱電機株式会社 | Digital watermarking device |
WO2006033953A1 (en) * | 2004-09-16 | 2006-03-30 | Thomson Licensing | Video codec with weighted prediction utilizing local brightness variation |
US20060150153A1 (en) * | 2004-12-04 | 2006-07-06 | Micah Altman | Digital object verification method |
GB2421133A (en) * | 2004-12-09 | 2006-06-14 | Sony Uk Ltd | Registering a water marked image by calculating distortion vector estimates |
US8020004B2 (en) | 2005-07-01 | 2011-09-13 | Verance Corporation | Forensic marking using a common customization function |
AU2006265780B2 (en) * | 2005-07-04 | 2012-06-28 | Commonwealth Scientific And Industrial Research Organisation | Image watermarking |
NZ565552A (en) * | 2005-07-04 | 2011-03-31 | Commw Scient Ind Res Org | Embedding bits into segmented regions of an image based on statistical features |
US8781967B2 (en) | 2005-07-07 | 2014-07-15 | Verance Corporation | Watermarking in an encrypted domain |
JP2009503948A (en) * | 2005-07-19 | 2009-01-29 | ベリマトリックス,インコーポレイティド | Hidden and robust mark for media identification |
FR2890103A1 (en) * | 2005-08-25 | 2007-03-02 | Snecma | Movable gas turbine engine blade e.g. movable high-pressure turbine blade, has air deflector positioned based on air flow that is centrifugal or centripetal, to project air circulating in cavity towards wall of cavity |
JP4124783B2 (en) * | 2005-08-30 | 2008-07-23 | 富士通株式会社 | Information embedding device and information embedding program |
CN1333371C (en) * | 2005-10-20 | 2007-08-22 | 李京兵 | Digital watermark method capable of resisting geometric attack and conventional attack |
CN100357971C (en) * | 2006-01-18 | 2007-12-26 | 李京兵 | Wavelet-based geometric attack resistant digital watermark method |
KR101213161B1 (en) * | 2006-06-14 | 2012-12-17 | 삼성전자주식회사 | Video watermarking apparatus in compression domain and method using the same |
US8107669B2 (en) * | 2006-06-14 | 2012-01-31 | Samsung Electronics Co., Ltd. | Video watermarking apparatus in compression domain and method using the same |
US8102863B1 (en) | 2006-06-27 | 2012-01-24 | Qurio Holdings, Inc. | High-speed WAN to wireless LAN gateway |
US8615778B1 (en) | 2006-09-28 | 2013-12-24 | Qurio Holdings, Inc. | Personalized broadcast system |
EP2095560B1 (en) | 2006-10-11 | 2015-09-09 | The Nielsen Company (US), LLC | Methods and apparatus for embedding codes in compressed audio data streams |
CN100421123C (en) * | 2006-10-16 | 2008-09-24 | 李京兵 | Robust two-value text image multi-watermark realizing method |
CN100421124C (en) * | 2006-10-16 | 2008-09-24 | 李京兵 | Method for local non-linear geometric attack resisting watermark based on image phase characteristic |
US7983440B1 (en) * | 2006-11-02 | 2011-07-19 | Qurio Holdings, Inc. | Selection of I-frames for client-side watermarking |
US7738676B1 (en) | 2006-11-02 | 2010-06-15 | Qurio Holdings, Inc. | Client-side watermarking using hybrid I-frames |
US7802306B1 (en) | 2006-11-30 | 2010-09-21 | Qurio Holdings, Inc. | Multiple watermarks for digital rights management (DRM) and content tracking |
US8000474B1 (en) * | 2006-12-15 | 2011-08-16 | Quiro Holdings, Inc. | Client-side protection of broadcast or multicast content for non-real-time playback |
US20080159533A1 (en) * | 2006-12-28 | 2008-07-03 | At&T Knowledge Ventures, Lp | System and method of processing data |
US8358703B2 (en) | 2007-03-23 | 2013-01-22 | Thomson Licensing | Modifying a coded bitstream |
JP5467651B2 (en) * | 2007-06-14 | 2014-04-09 | トムソン ライセンシング | Encoding bitstream modification |
JP4871841B2 (en) | 2007-11-08 | 2012-02-08 | キヤノン株式会社 | PRINT CONTROL DEVICE, PRINT CONTROL METHOD, PROGRAM THEREOF, AND STORAGE MEDIUM |
JP2009177470A (en) * | 2008-01-24 | 2009-08-06 | Fuji Xerox Co Ltd | Document management device and program |
JP5080304B2 (en) * | 2008-02-08 | 2012-11-21 | アヴァシス株式会社 | Display method of image data with confidential data inserted |
US8259938B2 (en) | 2008-06-24 | 2012-09-04 | Verance Corporation | Efficient and secure forensic marking in compressed |
US8320606B1 (en) | 2008-08-29 | 2012-11-27 | Adobe Systems Incorporated | Video watermarking with fast detection |
US9667365B2 (en) | 2008-10-24 | 2017-05-30 | The Nielsen Company (Us), Llc | Methods and apparatus to perform audio watermarking and watermark detection and extraction |
US8359205B2 (en) | 2008-10-24 | 2013-01-22 | The Nielsen Company (Us), Llc | Methods and apparatus to perform audio watermarking and watermark detection and extraction |
US8121830B2 (en) * | 2008-10-24 | 2012-02-21 | The Nielsen Company (Us), Llc | Methods and apparatus to extract data encoded in media content |
US8340343B2 (en) | 2008-10-31 | 2012-12-25 | General Instrument Corporation | Adaptive video fingerprinting |
US8385590B1 (en) | 2008-11-05 | 2013-02-26 | Adobe Systems Incorporated | Video watermarking with temporal patterns |
US8199969B2 (en) | 2008-12-17 | 2012-06-12 | Digimarc Corporation | Out of phase digital watermarking in two chrominance directions |
US9117268B2 (en) | 2008-12-17 | 2015-08-25 | Digimarc Corporation | Out of phase digital watermarking in two chrominance directions |
EP2404443A4 (en) | 2009-03-03 | 2013-09-04 | Digimarc Corp | Narrowcasting from public displays, and related arrangements |
CA3094520A1 (en) | 2009-05-01 | 2010-11-04 | The Nielsen Company (Us), Llc | Methods, apparatus and articles of manufacture to provide secondary content in association with primary broadcast media content |
US8213673B2 (en) * | 2009-06-09 | 2012-07-03 | Avio Technology, Inc. | Watermarking of motion pictures |
CN101662686B (en) * | 2009-09-07 | 2011-06-15 | 江南大学 | DCT domain video watermark treatment method based on texture features |
CN101847250B (en) * | 2010-04-27 | 2012-05-23 | 浙江工商大学 | Vector map data blind watermarking method based on DCT transformation |
US8838977B2 (en) | 2010-09-16 | 2014-09-16 | Verance Corporation | Watermark extraction and content screening in a networked environment |
JP5793865B2 (en) * | 2010-12-28 | 2015-10-14 | 富士通株式会社 | Digital watermark embedding device, digital watermark embedding computer program, digital watermark detection device, and digital watermark detection computer program |
CN102238388B (en) * | 2011-06-29 | 2013-03-20 | 西安电子科技大学 | Self-adaptive robust video watermarking method based on AVS (Audio Video Standard) |
US8615104B2 (en) | 2011-11-03 | 2013-12-24 | Verance Corporation | Watermark extraction based on tentative watermarks |
US8923548B2 (en) | 2011-11-03 | 2014-12-30 | Verance Corporation | Extraction of embedded watermarks from a host content using a plurality of tentative watermarks |
US8682026B2 (en) | 2011-11-03 | 2014-03-25 | Verance Corporation | Efficient extraction of embedded watermarks in the presence of host content distortions |
US8533481B2 (en) | 2011-11-03 | 2013-09-10 | Verance Corporation | Extraction of embedded watermarks from a host content based on extrapolation techniques |
US8745403B2 (en) | 2011-11-23 | 2014-06-03 | Verance Corporation | Enhanced content management based on watermark extraction records |
US9547753B2 (en) | 2011-12-13 | 2017-01-17 | Verance Corporation | Coordinated watermarking |
US9323902B2 (en) | 2011-12-13 | 2016-04-26 | Verance Corporation | Conditional access using embedded watermarks |
US9571606B2 (en) | 2012-08-31 | 2017-02-14 | Verance Corporation | Social media viewing system |
US8869222B2 (en) | 2012-09-13 | 2014-10-21 | Verance Corporation | Second screen content |
US9106964B2 (en) | 2012-09-13 | 2015-08-11 | Verance Corporation | Enhanced content distribution using advertisements |
US8726304B2 (en) | 2012-09-13 | 2014-05-13 | Verance Corporation | Time varying evaluation of multimedia content |
US9286912B2 (en) | 2012-09-26 | 2016-03-15 | The Nielsen Company (Us), Llc | Methods and apparatus for identifying media |
US9224184B2 (en) | 2012-10-21 | 2015-12-29 | Digimarc Corporation | Methods and arrangements for identifying objects |
US9262794B2 (en) | 2013-03-14 | 2016-02-16 | Verance Corporation | Transactional video marking system |
US9251549B2 (en) | 2013-07-23 | 2016-02-02 | Verance Corporation | Watermark extractor enhancements based on payload ranking |
CN103414892B (en) * | 2013-07-25 | 2016-08-10 | 西安空间无线电技术研究所 | The Image Hiding that a kind of Large Copacity is incompressible |
US9208334B2 (en) | 2013-10-25 | 2015-12-08 | Verance Corporation | Content management using multiple abstraction layers |
US9596521B2 (en) | 2014-03-13 | 2017-03-14 | Verance Corporation | Interactive content acquisition using embedded codes |
US10289856B2 (en) * | 2014-10-17 | 2019-05-14 | Spatial Digital Systems, Inc. | Digital enveloping for digital right management and re-broadcasting |
CN104539965B (en) * | 2015-01-12 | 2017-08-01 | 中国电子科技集团公司第三十八研究所 | A kind of H.264/AVC compressed domain video watermark insertion and extracting method |
US10902541B2 (en) | 2018-10-23 | 2021-01-26 | International Business Machines Corporation | Watermarking a digital image |
CN110390620B (en) * | 2019-07-12 | 2022-12-23 | 浙江工商大学 | Contour map data blind watermarking method based on curve Fourier transform |
CN110458747A (en) * | 2019-07-29 | 2019-11-15 | 淮阴工学院 | The reversible watermarking algorithm of robust based on invariant features |
US11501786B2 (en) | 2020-04-30 | 2022-11-15 | The Nielsen Company (Us), Llc | Methods and apparatus for supplementing partially readable and/or inaccurate codes in media |
CN113506580A (en) * | 2021-04-28 | 2021-10-15 | 合肥工业大学 | Audio watermarking method and system capable of resisting random cutting and dubbing |
CN114897659B (en) * | 2022-05-09 | 2023-12-29 | 南京师范大学 | Vector geographic data zero watermark generation method and zero watermark information detection method |
CN115278314B (en) * | 2022-07-08 | 2023-10-13 | 南京大学 | Multi-value digital video watermark embedding and blind extraction method |
Family Cites Families (38)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2196167B (en) * | 1986-10-01 | 1991-01-02 | Emi Plc Thorn | Apparatus for marking a recorded signal |
GB8806452D0 (en) * | 1988-03-18 | 1988-04-20 | Imperial College | Digital data security system |
US4939515A (en) * | 1988-09-30 | 1990-07-03 | General Electric Company | Digital signal encoding and decoding apparatus |
US5319735A (en) * | 1991-12-17 | 1994-06-07 | Bolt Beranek And Newman Inc. | Embedded signalling |
US5721788A (en) * | 1992-07-31 | 1998-02-24 | Corbis Corporation | Method and system for digital image signatures |
US5636292C1 (en) * | 1995-05-08 | 2002-06-18 | Digimarc Corp | Steganography methods employing embedded calibration data |
GB9400971D0 (en) * | 1994-01-19 | 1994-03-16 | Mor Limited | Method of and apparatus for manipulating digital data |
US5530751A (en) * | 1994-06-30 | 1996-06-25 | Hewlett-Packard Company | Embedded hidden identification codes in digital objects |
US5568570A (en) * | 1994-09-30 | 1996-10-22 | Eastman Kodak Company | Method and apparatus for reducing quantization artifacts in a hierarchical image storage and retrieval system |
US5646997A (en) * | 1994-12-14 | 1997-07-08 | Barton; James M. | Method and apparatus for embedding authentication information within digital data |
GB9500285D0 (en) * | 1995-01-07 | 1995-03-01 | Central Research Lab Ltd | A method of labelling an audio signal |
US5530759A (en) * | 1995-02-01 | 1996-06-25 | International Business Machines Corporation | Color correct digital watermarking of images |
GB9502274D0 (en) * | 1995-02-06 | 1995-03-29 | Central Research Lab Ltd | Method and apparatus for coding information |
US5659726A (en) * | 1995-02-23 | 1997-08-19 | Sandford, Ii; Maxwell T. | Data embedding |
GB9504221D0 (en) * | 1995-03-02 | 1995-04-19 | Mor Limited | Method of and apparatus for manipulating digital data works |
CA2218957C (en) * | 1995-05-08 | 2005-01-25 | Digimarc Corporation | Steganography systems |
US5613004A (en) * | 1995-06-07 | 1997-03-18 | The Dice Company | Steganographic method and device |
EP0766468B1 (en) * | 1995-09-28 | 2006-05-03 | Nec Corporation | Method and system for inserting a spread spectrum watermark into multimedia data |
US5859920A (en) * | 1995-11-30 | 1999-01-12 | Eastman Kodak Company | Method for embedding digital information in an image |
US5664018A (en) * | 1996-03-12 | 1997-09-02 | Leighton; Frank Thomson | Watermarking process resilient to collusion attacks |
US5970140A (en) * | 1996-05-08 | 1999-10-19 | The Regents Of The University Of California | Modular error embedding |
JP3982836B2 (en) * | 1996-07-16 | 2007-09-26 | コーニンクレッカ フィリップス エレクトロニクス エヌ ヴィ | Method for detecting watermark information embedded in an information signal |
US6282299B1 (en) * | 1996-08-30 | 2001-08-28 | Regents Of The University Of Minnesota | Method and apparatus for video watermarking using perceptual masks |
JP3262159B2 (en) * | 1996-09-04 | 2002-03-04 | 日本電気株式会社 | Spread spectrum watermark processing method |
US5848155A (en) | 1996-09-04 | 1998-12-08 | Nec Research Institute, Inc. | Spread spectrum watermark for embedded signalling |
US5809139A (en) * | 1996-09-13 | 1998-09-15 | Vivo Software, Inc. | Watermarking method and apparatus for compressed digital video |
US5915027A (en) * | 1996-11-05 | 1999-06-22 | Nec Research Institute | Digital watermarking |
JP3649537B2 (en) * | 1996-11-27 | 2005-05-18 | 日本アイ・ビー・エム株式会社 | Data hiding method and data extracting method |
US5875249A (en) * | 1997-01-08 | 1999-02-23 | International Business Machines Corporation | Invisible image watermark for image verification |
US6185312B1 (en) * | 1997-01-28 | 2001-02-06 | Nippon Telegraph And Telephone Corporation | Method for embedding and reading watermark-information in digital form, and apparatus thereof |
JP3534577B2 (en) * | 1997-07-23 | 2004-06-07 | 日本電信電話株式会社 | Information multiplexing method, information extraction method and their devices |
JP3560441B2 (en) * | 1997-04-07 | 2004-09-02 | 日本アイ・ビー・エム株式会社 | Multiple frame data hiding method and detection method |
US5940135A (en) * | 1997-05-19 | 1999-08-17 | Aris Technologies, Inc. | Apparatus and method for encoding and decoding information in analog signals |
US6108434A (en) * | 1997-09-12 | 2000-08-22 | Signafy, Inc. | Counteracting geometric distortions for DCT based watermarking |
JP4064506B2 (en) * | 1997-09-17 | 2008-03-19 | パイオニア株式会社 | Digital watermark superimposing method, detecting method and apparatus |
US6037984A (en) * | 1997-12-24 | 2000-03-14 | Sarnoff Corporation | Method and apparatus for embedding a watermark into a digital image or image sequence |
US6208745B1 (en) * | 1997-12-30 | 2001-03-27 | Sarnoff Corporation | Method and apparatus for imbedding a watermark into a bitstream representation of a digital image sequence |
US6252971B1 (en) * | 1998-04-29 | 2001-06-26 | Xerox Corporation | Digital watermarking using phase-shifted stoclustic screens |
-
1998
- 1998-07-17 US US09/118,467 patent/US6154571A/en not_active Expired - Lifetime
-
1999
- 1999-06-17 JP JP17104999A patent/JP3687725B2/en not_active Expired - Fee Related
- 1999-06-23 EP EP10182712.9A patent/EP2306726B1/en not_active Expired - Lifetime
- 1999-06-23 EP EP99111935A patent/EP0967803A3/en not_active Withdrawn
- 1999-06-23 EP EP10182728A patent/EP2306727A3/en not_active Withdrawn
- 1999-06-23 CA CA002276378A patent/CA2276378C/en not_active Expired - Fee Related
- 1999-06-23 EP EP10182704A patent/EP2306725A3/en not_active Withdrawn
-
2000
- 2000-04-26 US US09/558,728 patent/US6738493B1/en not_active Expired - Lifetime
- 2000-04-26 US US09/559,808 patent/US6278792B1/en not_active Expired - Lifetime
- 2000-04-26 US US09/558,727 patent/US6724911B1/en not_active Expired - Lifetime
-
2004
- 2004-05-31 JP JP2004162106A patent/JP3978613B2/en not_active Expired - Fee Related
-
2007
- 2007-02-05 JP JP2007026098A patent/JP4506991B2/en not_active Expired - Fee Related
-
2010
- 2010-03-15 JP JP2010057425A patent/JP4947172B2/en not_active Expired - Fee Related
-
2012
- 2012-01-16 JP JP2012006049A patent/JP5321698B2/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
JP5321698B2 (en) | 2013-10-23 |
EP2306726A3 (en) | 2012-09-05 |
JP3978613B2 (en) | 2007-09-19 |
EP2306727A2 (en) | 2011-04-06 |
JP4947172B2 (en) | 2012-06-06 |
US6278792B1 (en) | 2001-08-21 |
JP2004304834A (en) | 2004-10-28 |
US6154571A (en) | 2000-11-28 |
CA2276378A1 (en) | 1999-12-24 |
EP2306726A2 (en) | 2011-04-06 |
JP2010136456A (en) | 2010-06-17 |
EP0967803A3 (en) | 2001-10-10 |
US6724911B1 (en) | 2004-04-20 |
EP2306725A3 (en) | 2012-09-05 |
JP2000138818A (en) | 2000-05-16 |
EP0967803A2 (en) | 1999-12-29 |
JP3687725B2 (en) | 2005-08-24 |
EP2306725A2 (en) | 2011-04-06 |
EP2306727A3 (en) | 2012-09-05 |
JP2012085358A (en) | 2012-04-26 |
JP2007189708A (en) | 2007-07-26 |
US6738493B1 (en) | 2004-05-18 |
EP2306726B1 (en) | 2013-09-18 |
JP4506991B2 (en) | 2010-07-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CA2276378C (en) | Robust digital watermarking | |
US6108434A (en) | Counteracting geometric distortions for DCT based watermarking | |
EP0851679B1 (en) | Identification data insertion and detection system for digital data | |
AU741281B2 (en) | Method and apparatus for selective block processing | |
Podilchuk et al. | Image-adaptive watermarking using visual models | |
Swanson et al. | Multiresolution scene-based video watermarking using perceptual models | |
JP4027542B2 (en) | Digital watermarking method for digital objects | |
EP1164543B1 (en) | Digital information embedding/extracting | |
US6944313B1 (en) | Method and device for inserting and decoding a watermark in digital data | |
Piva et al. | A DWT-based object watermarking system for MPEG-4 video streams | |
CA2321034A1 (en) | Field-based watermark insertion and detection | |
JP2009514305A (en) | How to embed data in an information signal | |
WO2007032753A1 (en) | Coefficient selection for video watermarking | |
US6724913B1 (en) | Digital watermarking | |
WO2003003285A1 (en) | Method and system for watermarking an electrically depicted image | |
Zlomek | Video watermarking | |
GB2390502A (en) | Watermarking material using a bandwidth adapted codeword | |
KR19980080152A (en) | How to insert a digital watermark |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
EEER | Examination request | ||
MKLA | Lapsed |
Effective date: 20160623 |
|
MKLA | Lapsed |
Effective date: 20160623 |