WO2006021247A1

WO2006021247A1 - Method for detecting errors in address databases

Info

Publication number: WO2006021247A1
Application number: PCT/EP2005/006132
Authority: WO
Inventors: Werner Hautsch
Original assignee: Siemens Aktiengesellschaft
Priority date: 2004-08-21
Filing date: 2005-06-08
Publication date: 2006-03-02
Also published as: DE102004040600B3

Abstract

The invention relates to a method for detecting errors in address databases for address read systems of letter sorting machines comprising OCR units and video coding systems. According to said method, during reading, the video coding sequences of the images of letters rejected after video coding or the distribution codes of the erroneously sorted letters determined during sorting and distribution of the letters read in the address read system are stored and the rejections are counted in relation to the respective video coding sequence or the wrong distributions are counted in relation to the determined distribution codes. When a defined number is exceeded and lies significantly above a suitably selected lower threshold value, images of rejections including the respective coding sequences or the images recorded after wrong distribution and during reprocessing of the letters in a letter sorting machine are stored as address database errors and output for error correction.

Description

description

Method for determining errors in address databases

The invention relates to a method for determining errors in address databases for address reading systems of mail sorting systems with OCR units and video coding systems.

The systems used today for the automatic sorting of shipments contain reading systems "with the aid of which the ^' address of the respective program is automatically read and from this the distribution information necessary for the sorting is derived.

Addresses that can not be recognized by the OCR unit with the required security are then further processed in a video coding system. In this case, the electronic images of the programs are displayed on screens of the video decoding stations. The operators of these video coding platforms have the task of manually determining the information necessary for the distribution of the transmissions. This is not a copying of the address. Such a procedure is far too costly and would lead to a completely unsatisfactory cost situation. On the contrary, precise coding of the respective address structures establishes specific coding rules according to which only certain parts of the address must be entered. This informational extract is defined in such a way that the necessary distribution information can be obtained in most cases in a clear manner and only in a few cases further decisions by the surgeon are necessary. This method is known as extraction coding method.

Consignments in which no distribution information can also be determined by using video coding must be sorted out into a separate compartment of the sorting machine and then Send be further processed by hand sorting. This manual work causes high costs and should therefore be avoided as much as possible.

The reasons for the failure of reading systems with OCR reading and video encoding processes are manifold:

1. The shipment was inserted incorrectly:

The consignment was accidentally inserted into the material input of the sorting machine with the back to

Scanner was inserted so that the recipient address could not be detected.

Errors of this kind are easily recognizable and can be remedied by an¬ closing proper insertion without large additional costs.

2. The address is illegible:

In the electronic picture the address is not recognizable even for a human being. This may be due to an objec- tively poor writing or printing quality. In many cases, however, the reason lies in the image recording system. For very unfavorable contrast ratios, e.g. silver-colored writing on a dark background, even the high-speed scanners used today in sorting machines are not able to record a readable image of the address.

The addresses of broadcasts of this type can only be read on the original and must therefore be processed in the hand distribution.

3. The address is incorrect:

The address information on the shipment does not correspond to a real postal address. In many cases these are typographical errors identifiable by a person with a good knowledge of local distribution geography. can be decorated. In some cases, however, further investigations must be made, which often also require postal opening of the consignment.

4. Although the address is correct, it can not be identified in the address database used. To carry out the OCR reading and video coding process, a database must be available which contains all addresses on a nationwide basis and, in so doing, the distribution information for each address assigns, which is needed for sorting the respective program. This database is called address database or address dictionary. If certain addresses in this database are now missing or have errors, they can neither be read by the OCR reading unit nor encoded by means of a video coding system, even with optimum printing or writing quality. They must then be manually sorted by persons with good knowledge of the local distribution geography.

Another reason for the above situation is the unavoidable typing error when entering the video coding sequences, even with well-trained employees.

In the operational operation of a sorting center, defects in the address database are the most common reason for processing in manual sorting. However, these methods are difficult to recognize in the methods used hitherto the content of the Adressdaten¬ bank to be able to distinguish the shortcomings of the address database from the other reasons mentioned above. This decision can only be made by persons with good knowledge be made the local distribution geography and is ent¬ consuming expensive.

The invention has for its object to provide a low-cost method for detecting errors in address databases,

According to the invention, the object is achieved by the features of claim 1. An¬. During the read operation, the video coding sequences of the images of the mailing surfaces having the address details or the distribution codes of misdistributed mailings determined during the sorting and distribution operation of the mail read in the address reading system are stored after the video coding. Furthermore, the Rück¬ instructions, based on the respective video coding sequence, or the incorrect distributions are counted based on the determined distribution codes. If a defined number is exceeded which is significantly above a suitably selected lower limit value, images of rejections with the respective coding sequences or the images recorded in a mail sorting system after the mis-distribution during further processing of the mailpieces are stored as address database errors and subsequently displayed for error correction. By automatically identifying the database error, which both missing and inaccurate entries ^concerns', and the presentation of the rejected Videokodiersequenzen o of the Verteilkodes the misdirection with the images for subsequent correction can maintain the address database effortlessly up to date and the scope of Handor¬ tion be greatly reduced.

Advantageous embodiments of the invention are set forth in the dependent claims. In order to carry out the correction of the address database efficiently, it is advantageous if the representation of the video coding sequences or distribution codes characterizing the address database errors takes place with the associated images in an order corresponding to the address data bank entries.

It is also advantageous to record and display only a defined number of images for each video coding sequence or for each distribution code of an address database error. As a result, the storage and evaluation effort can be kept low.

Furthermore, it is advantageous to adjust the limit values controlling the recording of images dynamically in the course of a measurement, as a result of which the measurement duration can be selected almost arbitrarily in accordance with the existing conditions.

Subsequently, the invention will be explained in an embodiment with reference to the drawing.

Show

1 shows a frequency distribution of different rejected video coding sequences with the recording limit values,

2 shows a frequency distribution of the rejected video coding sequences according to list 1 with the recording limit values.

Missing or incorrect entries in the address database that are not appropriate for the addresses used make themselves noticeable in the same way during reading as during video coding. If z. If, for example, a street name in the address database is missing or misspelled, this leads to a rejection (both in reading and in video coding). tremfall to an error).

By analyzing the rejections of the video coding system, the shortcomings of the address database can thus be determined.

The following automatic procedure is used for this:

In operational operation, the coding sequence leading to the rejection is stored and counted for each rejection of the video decoding system.

If one plots the frequency of the rejections of the various video coding sequences, then a distribution is given in FIG. 1:

Over a continuum approximately uniformly distributed over all sequences, accumulation points for certain coding sequences are raised.

The reason for this distribution is as follows:

Typing and typing errors are i.a. statistically evenly distributed over all possible video coding sequences. There is no particular preference for particular sequences, while in the absence of an entry in the address database, each of them. Address directed transmission always causes a rejection of jewei¬ time coding sequence. The accumulation points are therefore caused by deficiencies of the address database. The absolute height of the accumulation point indicates the frequency of such addresses in the operational operation of the sorting system.

Since it is not possible to draw any conclusions about the causative addresses from the coding sequences alone, images of the relevant programs must additionally be stored. It is necessary to record the programs concerned in order to avoid a disorderly flood of data.

For this purpose, the image recording is controlled by two Grenz¬ values. The lower limit is set to be well above the continuum caused by typing and typing mistakes. He controls the beginning of the recording. First of all, no recording takes place. Only when the counter of a certain coding sequence exceeds the lower limit value is the electronic image of the consignment causing the rejection stored. This suppresses the recording of typing and typing rejections, and records only examples of real address database failures.

Conversely, it is unnecessary to record images for all rejections caused by a deficiency in the address database. Some examples are enough. For this purpose, a second, upper limit is introduced, after which the recording is suppressed again. The difference between the lower and upper limit value indicates the number of sample transmissions to be stored. It should i.a. suffice to record 5-10 examples each. The planned duration of the measurement must be considered when choosing the lower and upper limit values. The measurement must be stopped at the latest when the frequency of the randomly distributed continuum reaches the lower limit in the course of the measurement. Otherwise, the typing and writing errors would exceed the lower limit and lead to an undesirable storage load.

The size of the limits further determines the sensitivity of the measurement:

According to the rules of statistics, if a small lower limit is selected, only the very frequent shortcomings of the address from the continuum clearly stand out, while the fewer faults which frequently occur are not yet clearly enough inferred from the continuum.

In order to make them visible, the duration of the measurement and thus also the lower limit must be increased. It is also possible to automatically shift the limit values upward when the randomly distributed continuum comes close to the initially selected lower limit value. To this In this way, it is possible to set the duration of the measurement virtually as desired, and to detect defects which frequently occur in a single measurement and less frequently occurring defects at the same time.

The recorded examples can be arranged on the basis of the video coding sequences triggering them. This automatically results in a natural order of the recorded data material. Each shortcoming in the address database detected by the measurement is evidenced by its frequency and by a given set of examples, and may be determined by the

Postal code and the alphabetical order of the coded Na¬ men are ordered. The persons responsible for the maintenance of the address data thus receive a well-organized data material, which enables them to quickly recognize and supplement the missing or incorrect entries.

These facts are now explained in two concrete cases.

Case 1:

Assume that the task of video coding is to first specify the 5-digit postal code and then the first 3 letters of the street name.

For postal code 78315, the address database may be missing the entry "Rosenstraße" and the entry "Lilienstraße". Addresses of the kind

Mr. X.Y Mrs. Y.X

Rosenstr.15 ^• or Lilienstr.12

78315 Radolfzell 78315 Radolfzell

the through the sequences

78315 ROS or 78315 LIL be video-coded, then lead to a rejection of the Vi¬ deokodiersystems and thus to an increase in the relevant counter. If the lower limit 50 is exceeded, the image of the program is additionally stored. If the counter exceeds the upper limit of 53, the storage of images is again suppressed.

Typed coding sequences e.g. 78315 RLS or RPS are also counted, but they do not accumulate in this form, so that the limit is not exceeded and memory is not stored. The same applies to the sequence 78315 RAS, which results from the spelling error "Rasenstr"

After a certain period of time, the following is an example of the list of rejections for the postal code 78315:

List 1:

Videocoder-Sequeπz Counter Filenames of recorded images

LBL 21

LCL 14

LIL 140 LIL_51.jpg _, Lll__52.jpg _, LIL_53.jpg

LKL 11 - LPL 15

RAS 15

RLS 23

ROS 280 ROS_51.jpg, ROS_52.jpg, ROS_53.JPG RPS 14

RQS 13-

For the coding sequences LIL and ROS, which clearly exceeded the lower limit, the images of each 3 exemplary broadcasts were recorded, from which then the images in the ad- database were found missing addresses and in this ein¬ worn.

The graphical representation of the assignment of List 1 shows the course shown in FIG.

The method is suitable for determining all information contained in addresses but missing or incorrect information in the address database, provided that this information is also the subject of video coding, eg. B. -

• Postal Codes

• Place names

• street names • house numbers

• PO box numbers

• recipient name

Case 2:

Suppose the video coding task consists of entering the postcode, the first 3 letters of the street name and the house number. The address dictionary contains the "Rosenstraße", but only with the numbers 1-13, the numbers 15-19 are missing.

Then the following list assignment results:

List 2: Video Coding Sequence Counter Filenames of recorded images

RBS11 15

RLS13 13

ROS 15 105 ROS15_51.jpg, ROS15_52.jpg, ROS15_53 ROS 17 132 ROS17_51.jpg, ROS17_52.jpg, ROS17_53

ROS19 240 ROS19_51.jpg, ROS19_52.jpg, ROS19_53

RQS 19 13

The missing house numbers become clear, while the existing house numbers do not lead to an entry in the rejection list.

The method is also suitable for determining names feh¬ loin in _w address database, but frequently used by the public alternatives tivschreibweisen.

It is not tied to the existence of postcodes. If there are no postal codes, the place name serves as the order criterion.

The method can also be advantageously used in the evaluation of Fehlverteilun¬ gen. Misallocations are recognized during the fine distribution and delivery of the items.

They are then returned to the next distribution center for redistribution.

On this occasion, the items can be processed while running in a sorting system and images and the previous (wrong) distribution results can be recorded. If the frequency of the incorrect distributions is plotted as a function of the distribution code, the characteristic distribution according to FIG. 1 results again.

Systematic errors which indicate an error in the address database form accumulation points, while the incorrect distributions attributable to identification errors and typing errors are rather uniformly distributed. By appropriate choice of the lower and upper limit value, in turn, Belegex Emplare be recorded, which can be used for a targeted Improvement of the address database.

Claims

claims

1. A method for determining errors in address databases for address reading systems of mail sorting systems with OCR units and video coding systems, dadurchge ¬ indicates that in the course of the reading operation, the video coding sequences zu¬ after video encoding rejected images of the address information have the transmission surfaces or in the course the sorting and distribution operation of the read in the address reading system readings distribution codes of misdistributed broadcasts are stored and the rejections be¬ attracted to the respective video coding sequence or Fehl¬ distributions based on the determined distribution codes are counted that when a predetermined

Number, which is significantly above a suitably selected lower limit value, images of rejections with the respective coding sequences or the images recorded after misdistribution during reprocessing of the mailings in a mail sorting system are stored as address database errors and subsequently misled correction.

2. Method according to claim 1, characterized in that the representation of the video coding sequences or distribution codes characterizing the address database errors follows with the associated images in an order corresponding to the address database entries.

3. The method according to claim 1, dadurchgekenn ¬ records that by selecting an upper limit only a set number of images for each video coding sequence or for each distribution code of a Adressda- tenbankfehlers recorded and displayed.

4. The method according to claim 1 or 3, characterized in that the limit values controlling the recording of images are dynamically adjusted during the course of a measurement.