CA2179302A1

CA2179302A1 - Method of automated signature verification

Info

Publication number: CA2179302A1
Application number: CA002179302A
Authority: CA
Inventors: Mohamed Ali Moussa; Chih Chan
Original assignee: Quintet Inc
Current assignee: Individual
Priority date: 1993-12-17
Filing date: 1994-12-14
Publication date: 1995-06-22
Also published as: EP0737342A1; EP0737342A4; JPH09506730A; WO1995016974A1; AU1674795A; US5680470A

Abstract

A method of signature verification in which a test signature entered by an operator (204) may be preprocessed and examined for test features. The features may be compared against features of a set of template signature (206), and verified in response to the presence or absence of the test features in the template signatures (207). The test signature may be preprocessed, so as to normalize it and remove artifacts which are irrelevant to verification (205). The features of the template signatures may be stored in an associative memory or a data structure with associative memory capabilities (203).

Description

~ wo 95~l6974 2 i 7 9 3 ~ 2 PC~7594~14sss TITLE OF THE IN~'ENTION
11 , 12 ~ETHOD OF AUTOMATED SIGNATURE VERIFICATION

14 BA~,~OUNV OF THE INVENTION
16 1. Field of the Invention 18 This invention relates to signature verif ication .

20 2. Descri~tion of Related ~rt Z2 In a variety of applications, methods for verifying 23 that a person seeking access is in fact authorized are often 24 I equired. For example, keys are commonly used to require some ~.or~ of authentication before entry to a h~ ;n~ or a vehicle is 26 l~ermitted. l~here access is desired to software E~LOyLt ~ 3 or to 27 !;Oft~tlL~ ~,ol.~L~lled devices (such as an automated bank teller or 28 ~redi-- oard account) access is often verified by requiring that 21 79~Q
wo 9S/16974 PCT/USs4/l4588 1 the person seeking access enter a password or personal identifier 2 number ( "PIN" ), or by requiring that such inf ormation be recorded 3 on a magnetic stripe or in the memory of a "smart" card.

S While these methods of the prior art achieve the goal 6 o~ limiting access, they are subject to several drawbacks. tl) 7 PA .. ~ Is and PINs may be f orgotten, leading to persons who are 8 authorized but cannot achieve access. (2) Passwords and PINs are 9 often chosen without security cnncidorations in mind (they are 10 often chosen to be easily r~ ~-red), or are too short, leading 11 security systems which depend upon them to be subject to attack 12 by testing likely keys. (3) ~agnetic cards or "smart" cards may 13 be lost or stolen, leading to persons who can achieve access but 14 are not authorized .

16 It would be advantageous to have a method of 17 authorization which allows verif ication in response to the 18 signature of the person 6eeking access. However, known methods 19 of signature matching generally require costly human review of 20 the signature. Accordingly, it is an object of the invention to 21 provide a method of automated signature verification.

23 S~qMARY OF THE INVFNTION

The invention provides a method of automated signature 26 verification, in which a test signature, e.g., a signature 27 entered by an operator, may be ~Ld~ essed and ~YA~jn~d for test 28 features. The test features may be compared against features of WO 9S11697~ . 7 g 3 0 2 PC~/US9~/14~88 1 a set of template signatures, and veri~ied in response to the 2 presence or absence of the test features in the template 3 signatures. In a preferred embodiment, the test signature may be 4 prQprocessed, 50 as to n~ i 7e it and remove artifacts which are irrelevant to verification. In a preferred F~mho~ L, the 6 features of the template signatures may be cl~t~rm;n~cl and stored 7 irl an associative memory or a data structure with associative 8 ml~mory c~r~h;1ities, e.g., a discrete Hopfield ~rtificial neural 9 nf!twork. In a preferred ~nho~lir~ L, the method of verification m2~y be adjusted to greater or lesser sensitivity in response to 11 e~:ternal conditions.

Figure 1 shows an example system in which automated 16 s:ignature verif ication is used to control entry.

18 Figure 2 shows a process f low chart of a method of 19 alltomated signature verif ication .
21 Figure 3 shows a process flow chart of a method of 22 signature smoothing.

24 Figures 4A and 4B show a process flow chart of a method of signature rotation.

27 Figure 5 shows a process flow chart of a method of 28 ~ignature resizing.

wo 9S/1697~ 2 ~ 0 2 PCT/US94/14588 Figure 6 shows a process f low chart of a method of 2 signature feature extraction.

4 Figure 6A shows a process ~low chart of the step of identifying the pen movement feature.
7 Figure 6B shows a process f low chart of the step of 8 identifying the pen speed feature.

Figure 6C shows a process flow chart of the step of 11 identifying the pen status feature.

13 Figure 6D shows a process f low chart of the step of 14 identifying the pixel dispersion feature.
16 Figure 6E shows a process flow chart of the step of 17 identifying the e~ id~ In coordinate feature.

19 Figure 6F shows a process f low chart of the step of identifying the polar coordinate feature.

22 Figure 6G shows a process f low chart of the step of 23 identifying the stroke turning feature.

Figure 7 shows a process flow chart of a method of 26 signature feature storage (generation of a Hopfield weight 27 matrix).

wo 9S/1697~ 7 ~ 3 0 2 PCr/US94/14S88 Figures 8A and 8B show a process f low chart of a method 2 of' unweighted Levenshtein distance measure.

4 Figures 9A and 9B show a process f low chart of a method of2 signature comparison with template signatures.
7 Figure 10 shows a process f low chart of a method of 8 siLgnature accept/reject decision.

10 ~ DESCRTPTION OF THE PREFERRED EMBODIMENT

12 Figure 1 shows an example system in which automated 13 s ignature verification is used to control entry.

15 ~ A autoDIated signature verification system 101 may 16 c~mprise an input device 102 for receipt of an input signature 17 103, such as a writing implement 104 and a ~JL~:S:.ULe plate 105, 18 coupled to a processor 106 for receiving data relating to the 19 input signature 103. In a preferred ~"ho~;- L, the writing 20 ; l~ - lt 104 and EJL~S=.ULI~ plate 105 may comprise a 6tylus and 21 graphics tablet for freehand computer input, in which the 22 location of the stylus on the tablet and the pressure then 23 exerted are per~ l ly transmitted to the pLocessor 106, as are 24 known in the art (such as the "Acecat" graphics tablet made by 25 ~CECAD of Monterey, California). However, it would be clear to 2~ those skilled in the art, after perusal of this application, that 27 c~ther types of input device would also be workable, and are 28 Yithin the scope and spirit of the invention.

wo 95116974 ~' PC rlUss4l 14588 In a preferred embodiment, the processor 106 may 2 comprise a system having a processor, memory comprising a stored 3 program, memory comprising data, and input/output devices 107, as 4 is well known in the art. The operation and software structures 5 of this system are described herein in terms of their functions 6 and at a level of detail which would be clear to those of 7 ordinary skill in the art. It would be clear to anyone of 8 ordinary skill in the art, after perusal of this application, 9 that modification and/or plOyL ;n~ (using kno~n programming 10 techniques) of a processor of known design to achieve these 11 functions would be a straightforward task and would not require 12 undue experimentation. It would also be clear to those skilled in 13 the art, after perusal of this application, that processors of 14 other types could be adapted to methods shown herein without 15 undue experimentation, and that such other types of processor are 16 within the scope and spirit of the invention.

18 In response to the input signature 103, the processor 19 106~may generate a verification signal 108, which may be used to 20 verify the identity of the person writing the input signature 21 103. This verification signal 108 may be viewed by an operator, 22 may be coupled directly to a locking device 109, or may be 23 coupled to software within the processor 106 (or within another 24 processor). In a preferred Pmho~ nt, the verification signal 25 108 may be combined with other methods for verifying the identity 26 of the person, such as methods which are already known in the 27 art.

WO 9511697.1 2 1 7 9 3 0 2 PC~uS94~]4588 ~ ~... . .
SIGNATURE INPUT, STORAGE AND LATER VERIFICATION

3 Figure 2 shows a process f low chart of a method of . 4 a~ltomated signature veri~ication.

6 A method of automated signature verification may 7 ct~mprise a template input step 201, a template nnr~l;7ation step 8 2~2, a template storage step 203, a test input step 204, a test 9 n~rmalization step 205, a comparison step 206, and an accept/-r~aject decision step 207.

12 As shown herein, the method may use a system as 13 d,ascribed with reference to ~igure 1, and may proceed by 14 rlacording a set of template signatures, which are known to be v,~lid signatures, and may be used later for comparison with a 16 t~est signature. In a pre~erred ~nhor~ , the set of template 17 signatures may comprise at least five individual signatures.
18 These tQmplate signatures are each input and nor~-l i7r'd, and 19 stored for later comparison. The method may then proceed, when verification of a person's identity is desired, to verify a test 21 signature. The test signature is input and normalized, compared 22 with the template signatures, and accepted or rejected in 23 response to that comparison.

In the template input step 201, the person whose 26 identity is to be verified later may write a set of template 27 input signatures 103 on the input device 102, e.g., by 28 ~landwri~ing the input signature 103 with the writing implement ~` :

WO 95/16974 ~ PCr/uss4ll4s88 104 on the ~Ie~au~ plate 105. The template input signatures 103 2 are known to be valid signatures, and may be used later for 3 comparison with the test input signature 103.
At the template input step 201, the identity of the 6 person may be explicitly identified to the 6ystem, e.g., by mean6 7 of an additional input device 107, e.g., a text input device such 8 as a keyboard, or by means of other input devices such as a mouse 9 or other pointing device, voice input, photographic or other graphic input, or by other means of data input which are known in 11 the art. However, it is not required that the identity of the 12 person be explicitly identified, e.g., the system may compare a 13 test input signature 103 with all rec~rded template signatures, 14 and generate the verification signal 108 if there is a match with any stored set of template signatures.

17 In a preferred ~hoA;--r~, data transmitted by the 18 input device 107 to the processor 106 may be periodically 19 retrieved by the processor 106, as is well known in the art, and stored in a data structure associated with the template input 21 signature 103. In a preferred -Tnho~ , the data may comprise 22 a set of pixels, each of which may comprise a set of pixel data, 23 organized into a data structure as shown in the following table.
24 However, those skilled in the art would recognize, after perusal of this application, that other 6ets of data elements would be 26 workable, and are within the scope and spirit of the invention.

WO 95/lG974 ~17 ~ 3 ~ 2 PCT/US9~14!i88 P, -- { X" Y~, T" S~, PR, }
2 PZ = { X2 I Y2 I T2 l 52 ' PR2 }
3 P3 = { X3, Y3, T3, S3, PR3 }
4 * * *
6 Pn ~ Xn~ Yn~ Tn~ Sn~ PRn }
7 where Xj 2 X-coordinate of pixel, Y~ - Y-coordinate of 8 pixel, Ti = time when pixel :cLpL~d, Sj = pen-up/pen-down 9 6tatus of the writing implement 104 at that time, PR
lo pressure of the writing implement 104 at that time 1~
12 In the template normalization step 202, the processor 13 106 may convert the template input signature 103 into a 14 nnrr l j 7ed form. Use of a normalized form allows the processor 15 106 to remove features of t~Le template input signature 103 which 16 are deemed irrelevant to comparison with other signatures~ Such 17 irrelevant f eatures may include noise introduced by the input 18 device 102, orientation, and size.

In the template storage step 203, the template input 21 signatures 103 may be stored for later comparison with a test 22 s,ignature. In a preferred ~rho~;r--t, features of the template 23 input signatures 103 may be detP~m;n~d in response to the 24 t:emplate input signatUres 103 and stored in a manner which allows 25 ~Lssociative memory retrieval.

27 In the test input 5tep 204, the person whose identity 28 iLs to be verif ied may write a test input signature 103 on the g WO 95/1697J ~ 2~l 7~3 0 2 PCT/US94/14588 input device 102, in similar manner as the template input step 2 201. At the test input step 204, the identity of the person may 3 be explicitly identified to the system in similar manner as in 4 the template input step 108. At the test input step 204, the 5 processor 106 may capture si~ilar data as in the template input 6 step 201.

8 In the test normalization step 205, the processor 106 9 may normalize the test input signature 103 in similar manner as the template normalization step 202 is performed for the template 11 input signature 103.

13 In the comparison step 206, the processor 106 may 14 compare the test input signature 103 with the stored template input signatures 103. In a preferred D~nho~ rt, features of the 16 test input signature 103 may be determined with reference to the 17 test input signature 103 and compared with the stored template 18 input signatures 103 using associative memory retrieval.

In the accept/reject decision step 207, the processor 21 106 may determine whether to Yerify the person entering the test 22 input signature 103 in response to the comparison step 206.

WO 95116974 ~'~ 7 ~3 0 2 PCT~'USg.l/14!i88 SIGNATURE NORMAT r7ATION

3 In a preferred Pmho~ -nt, the template normalization 4 stlep 202 and the test normalization step 205 may each include a sml~othing step, a rotation step, and a resizing step.
7 Figure 3 shows a process f low chart of a method of 8 sil~nature smoothing.

It may occur that data captured by the graphics tablet 11 has a higher resolution than data which is transmitted to the 12 processor 106 , e. g ., because the graphics tablet has a higher 13 resolution than a graphics adapter used to transmit that data.
14 This may result in two pixels being mapped to the same location on the graphics adapter, which in turn may cause the input 16 signature 103 to appear not to be smooth, and may even cause 17 strokes of the input signature 103 to appear to zigzag.
18 Additionally, the manner in which the input signature 103 was 19 written may cause it not to be smooth.
21 ~ At an initialization step 301, a set of weights wl, wz 22 anjd U3~ and a threshold Cf, may be detPrminPd. A preferred value 23 fc,r v1 ~ay be 1, a preferred value for w2 may be 2, a preferred 24 value for U3 may be 1, and a preferred value for C~ may be 4.
26 At a smooth-once step 302, each pixel value for Xj and 27 Yj may be smoothed by computing a weighted average of that pixel, 28 it:s predPcPssor pixel, and its sUCcpcc~r pixel, as follows:

W0 95/1697~1 ~17 9 3 0 2 PCTIUS94114588 w1 Xj ~ + wz Xj + w3 X
2 new X j z -3 w1 + w2 + W3 w1 Yj 1 + wz Yj + w 6 new Y j =
7 w1 + w2 + W3 9 The 6mooth-once step 302 may be performed for each 10 pixel, except the first and last pixel. In a preferred ho~iir--~t~ each pixel value for new Xj and new Yj may be 12 computed using old values for that pixel, its pred~c~csnr pixel, 13 and its sUccF~c~nr pixel.

At a compute-distance step 303, a distance ~1I for each 16 pixel between the ~IC -_thed pixel values for Xj and Yj and the 17 post-smoothed pixel values for Xj and Yj may be computed. In a 18 preferred: '~o~ nt, this distance may be computed using a 19 standard ~lrl i d~n distance metric. The compute-distance step 20 303 may be performed for each pixel.

22 At a maximum-distance step 304, the maximum such 23 dist2nce for any pixel may be computed as follows:

2 5 d,N,~ = max { d j I 1 <= i <= n }

27 where n is the number of pixels.

~ WO 95/1697.1 ~ i7. 9 3 0 2 PCrtU594/14588 At a distance-threshold step 305, the maximum distance 2 ~""~ is compared with the threshold ~. If ~ is not greater 3 t~lan ~f, the method of smoothing is complete. Otherwise, the 4 m~-thod repeatedly assigns each pixel its new Xj and Yi values, alld then continues with the smooth-once step 3 02 . Because the 6 sIIooth-once step 302 causes the distance Cj between a new pixel 7 v.llue and its previous value to become smaller, the value d_,~
8 c~mputed at the ~-~i distance step 304 also becomes smaller, 9 ulltil it becomes smaller than the threshold Cf. Accordingly, the c~mparison at the distance-threshold step 305 will eventually 11 sllow ~ to be les6 than ~f, and the method of signature 12 smoothing will eventually terminate ( i . e., it will not proceed in 13 a~l "infinite loop").

Figures 4A and 4~ show a process flow chart of a method 16 oE signature rotation.

18 It may occur that the input signature 103 is written at 19 am angle from what would normally be expected, either due to p~asitioning of the graphics tablet, position of the person making 21 t~he input signature 103, or the manner in which the input 22 signature 103 was written. In a preferred ~mhodir-nt, the angle 23 of the input signature 103 is d~ectod and the input signature 24 103 is rotated to align it with a horizontal or vertical axis.
26 At an orientation step 401, the orientation of the 27 input signature 103 may be detPrm;n~. In a preferred 28 ~nho~lir t, the horizontal extent L,~ of the input signature 103 wo g5,l6974 ~ i 7 ~ 3 ~ 2 i `. PCT/U594/14S88 1 may be compared with the vertical extent Ly of the input 2 signature 103. If Lx >= Ly~ the input signature 103 is 3 dPt~rminP~' to be horizontal, otherwise the input signature 103 is 4 detPrminP~ to be vertical. The r---~in~lP~ of the method of signature rotation is described with reference to a horizontal 6 input signature 103, but treatment of a vertical input signature 7 103 would be clear to those skilled in the art after perusal of 8 this application.

At a partition step 402, the input signature 103 may be 11 partitioned into m equal intervals along the X axis. A preferred 12 value for m may be 64. The smallest and largest X coordinates 13 may be determined and the X interval for input signature 103 may 14 be divided into m equal intervals. Each pixel may be assigned to one of these intervals.

17 At an interval-statistics step 403, the mean Sj and 18 standard deviation 8j of the pixels in each interval j may be 19 computed, using known statistical formulae. The median T of the standard deviations 8j may be computed at step 404, using known 21 statistical formulae. The median absolute deviation ~D of the 22 standard deviations 8j from T may be computed at step 405, using 23 known statistical formulae.

At an outlier-detection step 406, outlying pixels are 26 removed. Those intervals for which (Sj - T) /I~AD exceeds a 27 threshold C are detPrmin~d to be outliers. A preferred value for 2 8 C may be 2 . 5 .

WO 9S116974 ~ Q 2 ` ~ PCT/VS94~14.S88 1 ] At a regression step 407, a regression line may be 2 computed f or non-outlier pixels, using known statistical 3 formulae. The re~ression line may have the form y = m x + }~, and 4 the angle of the slope regression line may be computed from its 5 slope, using known cJ~ L~ ic formulae.

7 ~ At a rotate step 408, each pixel of the input signature 8 103 may be rotated to a new X coordinate 2~nd Y coordinate 9 position, using known gP~ ~LiC formulae.

11 Figure 5 shows a process f low chart of a method of 12 signature resizing.

14 It may occur that the input signature 103 is written with a varying size. In a preferred ~ho~;t ~, the input 16 signature 103 is scaled to a uniform size.

18 At a r. - sctting step 501, a frame of size m by ~ is 19 s~lected to contain the input signature 103. A preferred value fc)r m may be 192. The orientation of the input signature 103, as 21 'li~t~nin~d at the orientation step 401.

23 At a horizontal-ratio step 502, a ratio r of the actual 24 s:Lze of the input signature to the frame size m may be computed, a-; follows:

2 7 r = Lx / m = .

WO 9S/1697.1 2 ~ 7 g ~ ~ ~ PCTII~S94114588 ~
At a scaling step 503, each pixel of the input 2 signature 103 may be scaled to a new X coordinate and Y
3 coordinate position, using known geometric formulae.
At a time-reset step 504, each pixel of the input 6 signature 103 may have its time Tj adjusted by subtracting the 7 start time of the input signature 103.

FEATURE EXTRACTION AND TENPLATE STORAGE
11 One aspect of the invention is the identif ication of 12 relatively constant features in signatures, which remain present 13 in the signature of a person even though that person' s signature 14 may be rewritten on differing occasions. One valuable indicator of the source of a person's signature is the strength of those 16 features identifiable in that person's template signatures.

18 One class of features may include time series data, 19 e.g_, pen-up/pen-down status, pen position, writing pressure, or writing speed or acceleration, each expressed as a function of 21 time. Another class of features may include parameters derived 22 from the input signature 103, e.g., number of strokes, total time 23 duration or duration of each stroke, number of pixels in each 24 interval (e.g., each interval identified in the partition step 402), centroid of all pixels, higher order moments, minimum and 26 maximum X and Y extent of the signature, or of an interval, peak 27 curvatures and locations thereof, starting location or direction, 28 ~nd otber aggr--gate value known In the art of stat ~ stics .

~ WO 95~16974 ~ ~ 7 9 3 ~ 2; PC~/US94/14588 -- In a preferred F~n~hoAi-- nt, each identified feature may 2 b(~ expressed as a vector of binary values, each etaualling "0" or 3 "1", i.e., a binary vector or bit string. This has the advantage 4 o~E reducing storage requirements.
6 Figure 6 shows a process f low chart of a method of 7 5 ignature f eature identif ication and representation .

9 In a preferred embodiment of the invention, seven specific features of the input signature 103 may be identified.
11 These features may include time series data, such as (1) v~ - L
12 of the writing implement 104 as a function of time, (2) speed of 13 the writing implement 104 as a function of time, (3) pen-up/pen-14 down status of the writing implement 104 as a function of time, and (4) pixel dispersion as a function of time. These features 16 ~lay also include time-in~Pr~nd~nt features, such as (5) a 17 ~ rl ;de~n coordinate map, (6) a polar coordinate map, and (7) a 18 ~;et of stroke turning positions in a ~llrl ;d~P~n coordinate map.

At a step 601, a pen ~ feature, comprising 21 llovement of the writing implement 104 as a function of time, may 22 I~e identified and represented as a bit vector.

24 At a step 602, a pen speed feature, comprising speed of the writing ;~rl~ -nt 104 as a function of time, may be 26 identified and represented as a vector of integers.

WO 95116974 2 i ~ .~ 3 ~ 2: PCTII~S94/14588 At a step 603, a pen status feature, comprising pen-2 up/pen-down status of the writing implement 104 as a function of 3 time, may be identified and represented as a bit vector.

At a step 604, a pixel dispersion feature, comprising 6 pixel dispersion as a function of time, may be identified and 7 represented as a bit vector.

9 At a step 605, a euclidean coordinate feature, comprising a Pllrl i~sln coordinate map of the input signature 103, 11 may be identified and Ie~L~seLed as an array of integers.

13 At a step 606, a polar coordinate feature, comprising a 14 polar coordinate map of the input signature 103, may be identif ied and represented as a vector of integers .

17 At a step 607, a stroke turning feature, comprising a 18 set of stroke turning positions in a euclidean coordinate map, 19 may be identif ied and represented as a bit vector.

21 Figure 6A shows a process flow chart of the step of 22 identifying the pen v L feature.

24 At a partition step 611, the input signature 103 may be 25 pArtitioned into a set of 21 bins of equal time duration. A
26 preferred value for M may be 32.

~16974 ~ PCT/US94/14~88 1 ¦ ' At a movement determination step 612, the total 2 movement of the writing implement 104 may be Cipt~rTn; hPd. In a 3 preferred ~mho~i- t, the sum 8 of the differences (Xj,1 - Xj~ may 4 be computed for each bin, where both pixels Xj and Xj,t belong to 5 the same bin.

7 At a quantization step 613, the sum is quantized by 8 setting a quantized result R to 1 if the sum ~ is negative, and 9 by setting 2 quantized result R to 0 if the sum 8 is nonnegative.
11 The feature may be represented by a vector of ~ bits of 12 t~le quantized result R.

14 Figure 6B shows a process f low chart of the step of iclentifying the pen speed feature.

17 At a partition step 621, the input signature 103 may be 18 p~lrtitioned into a set of ~ bins of equal time duration. A
19 p~-eferred value for M may be 32.
21 At a distance summing step 622, a su~ of P~o-lidr~n 22 distances of successive pixels in each bin may be determined, 23 e.g., a value for the each bin bti] is set to a sum of square-24 root( (X~ Xi)2 + (y~ _ yj)2 ), summed over all pixels ~Xj,Yj>
in that bin.

27 At a norr~ ;n~ step 623, each b[i] may be divided by 28 the total of all b[i].

wo 9511697~ ~ 1 7 g ~ ~t 2 ' ! j ~; Pcr/uss4ll4~88 The feature may be represented by a vector of 2 integers.

4 Figure 6C shows a process f low chart of the step of identifying the pen status feature.
7 At a partition step 631, the input signature 103 may be 8 partitioned into a set of }l bins of equal time duration. A
9 preferred value for M may be 32.
11 At a pen-up step 632, a status bit b[i] for each bin 12 may be set to "1" if any pixel in that bin has pen-up status.
13 Otherwise, the status bit b[i] for that bin may be set to "0".

The f eature may be represented by a vector of H bits .

17 Figure 6D shows a process f low chart of the step of 18 identifying the pixel dispersion feature.

At a partition step 641, the input signature 103 may be 21 partitioned into ~ set of lS bins of equal time duration. A
22 preferred value for ~I may be 32.

24 At a dispersion step 642, a standard deviation of X
coordinates sigma ~Xl) and a standard deviation of Y coordinates 26 sigma (Y~) may be t9t~t~rm;nPd, using known statistical formulae.

~ W0951~6974 ~1~93~2- PCTIU594114588 At an orientation step 643, a pixel dispersion bit b~i]
2 for each bin may be set to "0" if sigma(Xi) > sigma(Y~) and the 3 signature is horizontal, and may be set to "1" if sigma(Xj) ~
4 sigma(Yj) and the signature is detpnminpd to be horizontal. If S the signature i5 clptprm;np~l to be vertical, these "0" and "1" bit 6 values may be inverted.

8 The f eature may be represented by a vector of M bits .

Figure 6E shows a process flow chart of the step of 11 identifying the eUCl i~lPAn coordinate feature.

13 For this feature, the input signature 103 may be mapped 14 on~to an ~ x N matrix of bins b[x,y]. A preferred value for M may 15 be! 16; a preferred value for N may be 16.

17 At a centroid step 651, a coordinate <Xm,~n,Yn,~An> may be 18 m2~pped onto bin b[M/2,N/2].

At a mapping-ratio step 652, a farthest coordinate from 21 t~le center Xf for horizontal signatures (Yf for vertical 22 siLgnatures) may be detP~m;nPc~. A ratio r for mapping pixels may 23 b~ detPrm;nPd 2 5 I Xf - X",,~n r = l 26 1 ~/2 -WO 9S/16974 ~ ~ 7 9 $ 0 2 ~ PCT/U594/14S88 ~
At ~ mapping step 653, each pixel <Xk,Yk> may be mapped 2 onto a bin b [ i , j ], where 4 i = j k Xme~n r 6 j = I k Y nenn 7 r 8 At a pixel-count step 654, the number of pixels mapped 9 onto each bin b [ i, j ] may be counted .

11 At a norr~ ;n~ step 655, the number of pixels for 12 each bin b [ i, j ] may be divided by the total number of pixels in 13 the signature. In a preferred PmhO~; L, the normalized value 14 may be rounded up to the nearest integer if it comprises a 15 fractional value.

17 The feature may be ~ es~l.Led by an M x N array of 18 integers. In a preferred Pmho~ nt, each integer may be 19 represented by six bits in unsigned binary format, with values 20 greater than 63 represented by the bit string for 63, i.e., 21 "111111". When retrieved for Levenshtein distance comparison, 22 the binary data may be lnrarlrP~l into the H x 11 array of integers.

24 Figure 6F shows a process flow chart of the step of 25 identifying the polar coordinate feature.

27 For this feature, the input signature 103 may be mapped 28 onto a polar coordinate structure with H equidistant C~alOe~ r iC

~W095/16974 2l7`b~ 2 PcT/llS9~ 588 1 ri.ngs and N equiangular sectors in each ring. A preferred value 2 fc~r H may be 24; a preferred value for N may be 24.

4 At a centroid step 661, a coordinate <X",~.n,Ym"n> may be S ma~pped onto the origin of the polar coordinate system.
7 At a translation step 662, each pixel <Xk,Y~C> may be 8 t~anslated by subtraction of <Xm~n'Y~An>

At a mapping-ratio step 663, a farthest radius from the 11 c6!nter Rf may be detPrminPcl. A ratio r for mapping pixels may be 12 d6!tPrm; nP~

14 , Rf ¦
r= ' __ I M

17 At a ring-mapping step 664, each pixel <Xk,Yk> may be 18 m lpped onto a ring, where scluare-root ( X 2 + y 2 ring = k k 21 r 22 ~ At a sector-mapping step 665, each pixel <Xk,Y~> may be 23 m~pped onto a sector within its ring, where 25 theta = arctan (Yk/Xk), +27~ if needed to bring within (0,27~) 27 sector = theta ~ (180/7l) / (360/N) W095/16974 2~9~ ~ PCT/US94/14588 At a pixel-count step 666, the number of pixels mapped 2 onto each <ring, sector> tuple may be counted.

4 At a ring-summing step 667, the number o~ pixels for each ring may be summed, i.e., the number of pixels for all the 6 sectors in each ring are 6ummed and placed in H bins b [ i ] .

8 At a normalizing step 668, the number of pixels for 9 each bin b[i] may be divided by the total number of pixels in the signature. In a preferred ~rhg/l;. -nt, the normalized value may 11 be rounded up to the nearest integer if it comprises a fractional 12 value.

14 The feature may be represented by an M x N array of integers for bins b[ring,sector]; the value for the origin of the 16 polar coordinate may be discarded for this feature. In a 17 preferred ~ho~ir-nt, each integer may be represented by six bits 18 in unsigned binary format, with values greater than 63 19 represented by the bit string for 63, i.e., "111111". When retrieved for Levenshtein distance comparison, the binary data 21 may be llnr~l-k~d into the N x N array of integers.

23 Figure 6G shows a process flow chart of the step of 24 identifying the stroke turning feature.
26 For this feature, the input signature 103 may be mapped 27 onto an N x N matrix of bins btx,y]. A preferred value for M may 28 be 16; a preferred value for N may be 16.

~ wo gS/16974 ~17~3 0 2 PCT/I~S94J]4588 A centroid step 671 and a mapping-ratio step 672 may be 2 performed in like manner as the centroid step 651 and mapping-3 ratio step 652.
At a mapping step 673, each pixel <Xk,Yk~ which r~ay 6 cc,mprise a 6troke-turning point may be mapped onto a bin b[i,;], 7 in, like manner as pixels are mapped onto a bin b[i,j] in the 8 mapping step 653. As used herein, a stroke-turning point is a 9 pc~int where there is a change in stroke direction.

11 ~ In a preferred ~ nt, stroke-turning points may be 12 r~co~n;7Pd by ~YA~n;nin~ each set of five consecutive pixels for a 13 c~lange of direction as the middle point. A change of direction 14 m21y be recogn;7ed in a variety of ways, e.g., by determining if t~e middle point is a minimum, maximum, or inflection point using 16 krlown methods of elementary cAlc~ c~ applied to discrete points.

18 A pixel-count step 674 and a norr~l;7;n~ step 675 may 19 b~ performed in like manner as the pixel-count step 654 and the n~ i7~in~J step 655.

22 The feature may be ~ se.. Led by an 11 x N array of 2 3 imtegers . In a pref erred embodiment, each integer may be 24 r,~L~ser.Led by six bits in unsigned binary forr4at, with values 25 greater than 63 represented by the bit string for 63, i.e., 26 "111111". When retrieved for Levenshtein distance comparison, 27 the binary data nlay be l~nrArk~ into the }I x N array of integers.

wo 9~116974 ~ ~ 7 9 3 0 2 ` ~ - Pcr/uss4/14s88 FEATURE RETRIEVAL AND SIGNATURE COMPARISON

3 Once a f eature has been identif ied and represented as a 4 template f eature bit string, a weight matrix may be generated according to the discrete 80pf ield a2,y~ 1-L UlloUs network paradigm .
6 The discrete 80pf ield asynchronous network i5 known in the art 7 and 50 is not disclosed in detail here. A more complete 8 discus6ion may be found in "Neural Networks and Physical Systems 9 with r g~l-L Collective Computational Abilities", by John J.
80pfield, pllhli ~h.od 1982 in "Proceedings of the National Academy 11 of Sciences, U.S.A. 1979", pages 2554--2558, hereby inCOL,uuL~lted 12 by reference as if fully set forth herein.

14 Figure 7 shows a process flow chart of a method of signature feature storage (generation of a llopfield weight 16 matrix).

18 At a bipolar cu~lv~ iOn step 701, each binary value (O
19 or 1) may be converted to a bipolar value (-1 or +1), by replacing all "O" values with -1 values.

22 At an outer-product step 702, an outer product may be 23 computed for each binary vector with its transpose. Where the 24 binary vector is length m, the product ~ will be an m x m bipolar matrix.

~ WO 95116974 ~ 1 7 9 3 0 Z P~ V594/14588 At a summation step 703, the outer products for the 2 selected feature of all (five) of the template signatures are 3 added to generate a summed matrix M'.
At a zero-diagonal step 704, the main diagonal of the 6 su~med matrix 11' is set to zero . The resulting matrix M ' is 7 herein called a weight matrix or a memory matrix.

9 Once a feature in the test input signature 103 has been identified and represented as a test i'eature bit string, 11 differences between the test feature bit string and the template 12 feature bit strings may be rletPrm;np~ according to the 13 a~y~ onous update paradigm of the discrete Hopf ield network .
14 String distance may be computed according to the Levenshtein distance. The Levenshtein distance is known in the art and 80 is 16 na~t d; ~closP~l in detail here. A more complete discussion may be 17 fclund in "Binary Codes Capable of Correcting Deletions, 18 Inlsertions, and Reversals", by V.J. Levenshtein, pllhl;~hpd 1965 19 irl "Doklady AkAllP~;; Nauk SSR" 163(4), pages 845--848, hereby irlcv~ ted by reference as if fully set forth herein.

22 Figures 8A and 8B show a process flow chart of a method 23 oi unweighted Levenshtein distance measure.

Briefly, the Levenshtein distance between two strings 26 oi~ symbols may be computed by dPtP~m;n;n~ how many symbols must 27 be added, how many deleted, and how many substituted, from a 28 f:irst string A to a second string ~.

woss/l6974 ii~ ~ b i I PcrNS94114~88 At an initialization step 801, the length L. of string 2 A, the length Lb f string B, the deletion cost D, the insertion 3 cost I, and the substitution cost 8 may be determined. A
4 preferred value for D may be 1, a preferred value for I may be 1, 5 and a preferred value for 8 may be 1. A distance matrix M may be 6 allocated with an initial value for M(0,0) of zero.

8 In a deletion-loop 802, the value for N(i,o) may be set 9 to N(i-1,0) + D, for each value of i < LThe variable i may be 10 a counter.

12 In an insertion-loop 803, the value for ~(o,j) may be 13 set to M(0, j-l) + I, for each value of j < L~. The variable j 14 may be a counter.

16 In a substitution-loop 804, the counter variables i and 17 j may be allowed to range over each value i <= I,. and j <= Lb.
18 At step 805, the ith position of string A may be compared with 19 the ~th position of string B. For each location in N, the value for N(i,j) may be set to M(i-l,j-l) (at step 806), plus 8 if the 21 values of the strings differ in their ith and jth positions (at 22 step 807). At step 808, the value for N(i,j) may be computed for 23 achieving the substitution by deletion or insertion instead. At 24 step 809, the value for M(i, j) may be set to a lower value if one could be achieved by deletion or insertion instead.

27 Figures 9A and 9B show a process f low chart of a method 28 o~ Li~ture co~pA~iLon with ~e~pl--te Li~n~tures.

~179~1~2 WO 9S116974 ' : - PC~NS94/145X8 At an initialization-step 901, the weight matrix IS' and 2 a feature bit string b are known. An altered feature bit string 3 b ' may be set equal to b .
In an update-loop 902, a vector c may be computed as 6 t~le product IS' x b'. Each bit of b' may be altered to equal "1"
7 i~ that element of c is >= O, and otherwise may be altered to 8 ec~ual "O".

1 At a loop-complete step 903, the updated bit string b' 11 mzly be compared with the original bit string b. If they are 12 diLfferent, the updated bit string b' may be assigned to the 13 o~-iginal bit string b at step 904 and the update-loop 902 may be 14 r(~-entered. Otherwise, the update-loop 902 is complete and the m~thod may continue with step 905.

17 At a mea2,uL~ step 905, the Levenshtein distance of 18 t]le updated bit string b' from the template feature bit strings 19 m~y~be computed.
21 At a m;n; distance step 906, the minimum Levenshtein 22 distance computed may be determined to be the distance of the 23 test feature bit ~tring from the template feature bit strings.

In a preferred Pmho~ nt, the pen movement feature 26 data may comprise 32 bits, and may be packed into 4 bytes at 8 27 bits per byte. The pen speed feature data may comprise 32 28 integers, encoded with six bits per integer, thus 192 bits, and Wo 95116974 2 ~ 7 9 3 ~ 2 PCrlUS94/14588 1 may be packed into 24 bytes at 8 bits per byte. The pen status 2 feature data may comprise 32 bits, and may be packed into 4 bytes 3 at 8 bits per byte. The pixel dispersion feature data may 4 comprise 32 bits, and may be packed into 4 bytes at 8 bits per 5 byte. The e~ ;dP~n coordinate feature data may comprise 16 x 16 6 = 256 integers, encoded with six bits per integer, thus 1536 7 bits, and may be packed into 192 bytes at 8 bits per byte. The 8 polar coordinate feature data may comprise 24 integers, encoded 9 with six bits per integer, thus 144 bits, and may be packed into 10 18 bytes at 8 bits per byte. The stroke turning feature data may 11 comprise 16 x 16 = 256 bits, and may be packed into 32 bytes at 8 12 bits per byte.

14 In a pref erred Pmhorl i r L, a packed data structure f or 15 feature data may be expressed as follows in the C ~uyLaL_lng 16 language:

18 typedef struct 19 char movement [ 5 ]
char speed [ 2 5 ~
2 0 char status [ 5 ], char di spers i on [ 5 ];
21 char el~ P~n - map[193];
char polar _ map [ 19 ]
22 char turning _ position[33 ]
} PACKED _ FEATURE

In a pref erred Pmho~ , a packed data structure of five template signatures may be ~-L~IeSSed as follows in the C
~L UyL " .I~...ing language:

PACRED_FEATURE packed_template_f eature [ 5 ];

~ ~WO 9S116974 2 ~ 7 9 3 n 2 ~ - ` PcrNss4/l4s88 In.a preferred Pmho~i -nt, this data structure may be 2 st.ored in a database of signatures.
4 In a pref erred P~ho~; r ~ ~ an unpacked data structure fc,r feature data may be expressed as follows in the C l~LU~L ;nq 6 la.nguage:

8 typedef struct 9 char - v~ ~_[33];
short speed [ 3 2 ];
10 char status [ 3 3 ];
char disperSint33 ];
11 short o~ n map[256];
short po l ar _ map [ 2 4 ];
12 char turning _ position[257]
} UNPACKED FEATURE

14 In a preferred ~mho~ t, an llnr~ ed data ~-LU~.LUl~
15 of' f ive template signatures may be expressed as f ollows in the C
16 P1~O~L inq language:

18 UNPACKED_FEATURE l~nr~ Pd template_feature[5];

2 0 In a pref erred Pmhotl; r L, thi5 data structure may be 21 u~;ed for storing features which. are detPrm;"ecl for template 22 s ignatures and llnra~lrPd from a database of signatures.

24 An example Levenshtein distance comparison is shown in 25 tlle following table for two 20-bit bitstrings:

' Wo 95116974 ~! 1 7 9 ~ ' PCrNS94tl4588 1 Let a test feature vector of 20 bits be defined as follows:
~'11000 10110 11101 10101"
2 Let a template feature vector of 20 bits be defined as follows:
"01101 01011 01110 11010"

2 2 1 1 2 3 4 5 6 7 8 9 10 11 12 i3 14 15 16 17 18 13 19 18 17 16 15 14 13 12 11 19 10 9 _8 8 7 6 6 5 4 3 2 The Levenshtein distance between the two f eature 16 vectorF: is the last element, i.e., the lower right corner 17 element, which is 3.

19 Figure 10 shows a process f low chart of a method of 2 0 s ignature accept/ re j ect decis ion .

22 At a threshold step 1001, an acceptance threshold for 23 each feature is determined.

In a preferred ~n~ho~l;r--lt~ the acceptance threshold may 26 be der~rm;n~d in response to the distances between pairs of the 27 template input signatures 103. In a preferred ~mho~;r-nt with WO 95/16974 2 ~ 7 9 3 n ~ 1 ~ PCT/US94/14~i88 1 five template input signatures 103, there will be ten such 2 pairwise distances, as 6hown in the following table:
4 ~ 1 2 3 4 5 6 1 1 0 dl d2 d3 d4 7 2 I dl 0 d5 d6 d7 8 3 I d2 d5 0 d8 d9 9 4 I d3 d6 a8 0 dlo ~ 5 I d4 d7 d9 dlO 0 11 .
12 In a preferred ~r~o~';r-~t, these ten pairwise distances 13 m,ay be arranged in decreasing order, and the lcth distance may be 14 slelected as an acceptance threshold. A preferred value for lc may b,e 7 , i . e ., the 7th greatest distance may be selected as an 16 a,-ceptance threshold. The greater lc is, the tighter the 17 a~-ceptance threshold; i.e., when 3c is 9, the 9th greatest 18 distance may be selected as an acceptance threshold.

At an averaging step 1002, an average acceptance 21 t~hreshold ~ and an average distance 1~ may be computed.

23 At a comparison step 1003, the average acceptance 24 t]hreshold ~ and an average distance ~ may be compared. If the average acceptance threshold ~ is less than the average distance 26 ~, the test input signature 103 may be accepted and the 27 v~erification signal 108 may be generated. Otherwise, the test WO 95/16974 2 ~ ~ 9 3 0 2 PCTIUS94114588 ~
1 input signature 103 may be rejected and the veri~ication signal 2 108 may be ~bsent.

4 ~l ternative Emb 6 ~hile preferred r~-nho-l;r-nts are disclosed herein, ~any 7 variations are poqq; hl~ which remain within the concept and qcope 8 o~ the invention, and these variations would become clear to one 9 of ordinary skill in the art after perusal of the ~pecification, drawings and claims herein.

Claims

1. A method of automated signature verification, said method comprising receiving a plurality of first signals, each, representative of a template signature;
transforming each said first signal into a first structure representative of said template signature;
normalizing each said first structure, responsive to spatial features of said first structure, to generate a first normalized signature;
defining a first set of signature features responsive to each one of said plurality of first normalized signatures;
receiving a second signal, representative of a test signature;
transforming said second signal into a second structure representative of said test signature;
normalizing said second structure, responsive to spatial features of said second structure, to generate a second normalized signature;
defining a second set of signature features responsive to said second normal-ized signature;
comparing said second set of signature features with each one of said recorded first set of signature features; and verifying said test signature responsive to a result of said step of comparing.

2. A method as in claim 1, wherein said step of comparing comprises the step of computing a Levenshtein distance.

3. A method as in any previous claim, wherein said step of comparing comprises selecting a set of pairs of said first normalized signatures, computing a measure of distance for each said selected pair; and selecting an acceptance threshold, responsive to said computed measures of distance for said first normalized signatures;
for said second normalized signature, computing said measure of distance from each said first normalized signature;
defining a minimum distance for said computed measures of distance for said second normalized signature; and determining whether said minimum distance exceeds said acceptance threshold.

4. A method as in claim 3, wherein said step of selecting an acceptance threshold comprises sorting said computed measures of distance for said first normalized signatures into by magnitude, to generate an ordered list of computed measures;
selecting one said computed measure in said ordered list, responsive to a posi-tion on said ordered list for said computed measure; and defining said acceptance threshold responsive to said selected one said com-puted measure.

5. A method as any previous claim, wherein said step of recording said first set of signature features comprises the step of storing a bit vector associated with said signature features in an associative memory or in a data structure with associative memory capabilities; and wherein said step of comparing comprises the step of retrieving at least a part of said bit vector from said associative memory or data structure with associative memory ca-pabiiities.

6. A method as in any previous claim, wherein said step of recording said first set of signature features comprises the step of defining a set of weights for a discrete Hopfield artificial neural network weight matrix, responsive to a bit vector associated with said signature features; and wherein said step of comparing comprises the step of retrieving at least a part of said bit vector from said discrete Hopfield artificial neural network weight matrix.

7. A method as in any previous claim, wherein said step of defining a first set of signature features comprises constructing a binary vector for a first signature represent-ing said signature features for said first signature, wherein said binary vector constitutes less than about 300 bits.

8. A method as in any previous claim, wherein said first structure com-prises a plurality of pixels, each said pixel having spatial information associated therewith, and wherein said step of normalizing each said first structure comprises for a first said pixel, defining at least one second pixel as a neighborhood for said first pixel;
modifying said spatial information associated with said first pixel, responsive to said spatial information associated with at least one pixel in said neighborhood;
computing a distance measure for said first pixel, responsive to at least one pixel in said neighborhood; and repeating said steps of defining and modifying, responsive to said distance measure.

9. A method as in any previous claim, wherein said first structure com-prises a plurality of pixels, each said pixel having spatial information associated therewith, and wherein said step of normalizing each said first structure comprises dividing said first structure into a plurality of segments;
computing a measure of dispersion for each said segment;
defining, for a first said segment, a set of retained pixels less than all pixels in said first segment, responsive to said measure of dispersion for said first segment;
computing a regression line responsive to said retained pixels; and rotating said first signature responsive to said regression line.

10. A method as in any previous claim, wherein said first structure com-prises a plurality of pixels, each said pixel having spatial information associated therewith, and wherein said step of normalizing each said first structure comprises scaling said spatial infor-mation uniformly for the entire said first structure.

11. A method as in any previous claim, wherein said first structure com-prises a plurality of pixels, each said pixel having writing information associated therewith, and wherein said step of defining a first set of signature features comprises dividing said first structure into a plurality of segments;
computing a summary measure for each said segment, responsive to a set of pixels in said segment;

quantizing said summary measure for each said segment to generate a quan-tized measure, wherein said quantized measure comprises fewer than nine bits; and constructing a bit vector comprising said quantized measures;
wherein said writing information is in the set composed of a set of pen-up/pen-down status bits, at known time intervals, of said writing implement;
a set of measures of pixel dispersion, at known time intervals, of said first set of signatures;
a euclidean coordinate map of at least one of said first set of signatures;
a polar coordinate map of at least one of said first set of signatures; and a set of stroke turning positions in a euclidean coordinate map of at least one of said first set of signatures.

12. A method as in any previous claim, wherein said step of identifying a second set of features comprises recording data about at least three features in the set com-posed of a set of pen-up/pen-down status bits, at known time intervals, of said writing implement;
a set of measures of pixel dispersion, at known time intervals, of said first set of signatures;
a euclidean coordinate map of at least one of said first set of signatures;
a polar coordinate map of at least one of said first set of signatures; and a set of stroke turning positions in a euclidean coordinate map of at least one of said first set of signatures.

13. A method as in any previous claim, wherein said step of comparing comprises inputting said second set of features to an artificial neural network said artifi-cial neural network comprising a set of weights generated in response to said recorded data;
and generating an output from said artificial neural network.

14. A method as in any previous claim, wherein said step of comparing comprises inputting said second set of features to an artificial neural network, said artifi-cial neural network comprising a set of weights generated in response to a Hopfield weight matrix; and generating an output from said artificial neural network.

15. A method as in any previous claim, wherein said step of comparing comprises the step of generating a difference value in response to a set of differences between said first and second set of features;
said step if generating comprises the step of determining a difference threshold in response to said first set of features, and the step of comparing said difference value with said difference threshold.

16. A method as in any previous claim, wherein said step of verifying com-prises the steps of associating a set of authorization privileges with at least one said template sig-nature;
identifying said test signature with at least one said template signature; and verifying said test signature responsive to said set of authorization privileges associated with said at least one template signature.

17. A method of smoothing an input signature, comprising the steps of determining a time sequence of points in response to said input signature;
altering each point in response to a predecessor point and a successor point, for substantially every point in said time sequence;
determining a measure of smoothness of said time sequence after said step of altering; and repeating said step of altering in response to said measure of smoothness.

18. A method for rotating an input signature to a known axis line, com-prising selecting a set of pixels to represent an input signature, which set of pixels con-sists of fewer than all the pixels of the signature;
generating a line representing an axis of said selected set of pixels;
determining an angle between said line and a known axis line; and rotating all the pixels of said signature through said angle.

19. A method as in claim 18, wherein said step of selecting comprises dis-carding statistical outlier pixels.

20. A method as in any one of claims 18-19, wherein said step of generating a line comprises regression over said selected set of pixels.

21. A method as in any one of claims 18-20, wherein said step of selecting comprises allocating the pixels of the signature into a set of bins;
determining a statistical measure of pixels in each such bin;
selecting a subset of such bins in response to said statistical measures.

22. A data structure for recording a set of features of a written signature, comprising means for recording at least four features in the set composed of a set of pen-up/pen-down status bits, at known time intervals, of said writing implement;
a set of measures of pixel dispersion, at known time intervals, of said first set of signatures;
a euclidean coordinate map of at least one of said first set of signatures;
a polar coordinate map of at least one of said first set of signatures; and a set of stroke turning positions in a euclidean coordinate map of at least one of said first set of signatures.

23. A method of determining whether a first person is a particular individ-ual, said method comprising creating and storing signature features representing selected characteristics of a plurality of template signatures for the particular individual, said template signatures having been normalized responsive to spatial features thereof;
providing information representing said selected characteristics of a test signa-ture made on a signature transducer by said first person, said test signature having been nor-malized responsive to spatial features thereof;
comparing said selected characteristics of said test signature with said selected of each one of said template signatures; and generating a signal indicating a match if said selected characteristics of said test signature and said selected characteristics of said template signature match each other to a se-lected degree.