Method and apparatus for verification of signatures

• US 6,424,728 B1
• Filed: 12/02/1999
• Issued: 07/23/2002
• Est. Priority Date: 12/02/1999
• Status: Expired due to Term

• Current Assignee: Maan Ammar
• Original Assignee: Maan Ammar
• Inventors: Ammar, Maan
• Primary Examiner(s): Boudreau, Leo
• Assistant Examiner(s): Werner, Brian P.
• Application Number:
  US09/453,730
• Time in Patent Office: 964 Days
• Field of Search: 382/116, 382/119-123, 382/173, 382/197, 382/198, 382/206, 382/195, 382/216, 382/218, 340/5.52, 340/5.53, 340/5-1-, 356/71, 902/3-6, 434/155, 348/161
• US Class Current: 382/119
• CPC Class Codes: G06V 40/30 Writer recognition; Reading and verifying signatures

Abstract

An automatic signature verification system is set forth that utilizes a main routine for comparing signatures using forensic hand writing methodology. A secondary program is used to modify the algorithms used by the main program for making adjustments thereto based on either additional data consisting of a plurality of genuine or authenticated signatures or changes in a persons signature due to aging or some other physical change resulting in a change in signature features. Over seventy thousand signatures can be verified in one hour using a single personal computer. In addition, authentic signature data having a resolution as low as 80 dpi can be used.

Claims

1. A system for automatic signature authentication comprising:

a) a computer-implementable signature processing program to examine a digitized signature image obtained from a document, optionally a check, the signature image purportedly being the signature of a specific person, to generate an authentication decision authenticating the signature as genuine or rejecting the signature as a forgery;

b) an authenticated signature database of signature features for use by the signature processing program in generating the authentication decision; and

c) an evaluation program to evaluate the results produced by the signature processing program and to provide update information to the signature database for use by the signature processing program regarding examined signatures accepted as genuine by the signature processing program;

wherein;

d) the digitized signature image is scanned from a paper document;

e) multiple versions of the signature image are electronically generated and examined;

f) the results of examination are statistically compiled to enhance the authentication decision;

g) each version of the signature image is defined into segments, signature features are extracted from individual segments and the extracted signature features are evaluated against training data received from the evaluation program; and

h) the extracted signature features are each analyzed globally for the complete signature and locally for each segment.

2. A system according to claim 1 wherein the system comprises a global signature feature database including samples of signature features of genuine and forged signatures of persons other than the specific person'"'"'s signature together with available genuine samples of the specific person'"'"'s signature wherein the global signature feature database is employed to analyze the extracted signature features for the complete signature and the extracted signature features are analyzed locally for each segment utilizing at least one genuine signature of the specific person.

3. A system according to claim 2 wherein the global signature database is employed for signature recognition until the database includes a desired number of genuine samples of the specific person'"'"'s signature.

4. A system according to claim 2 wherein the digitized signature image is cleaned prior to generation of multiple versions by removing printed background characters, noise and unconnected lines not meeting signature image component criteria.

5. A system according to claim 2 wherein the multiple versions of the signature image comprise an unmodified version, a thinned version and a boundary-detected version.

6. A system according to claim 5 wherein the unmodified version is obtained by cleaning the signature image and is passed directly to a dissimilarity measurement procedure and wherein a second version of cleaned signature image is processed in a normalization procedure to adjust the image to a system standardized configuration with regard to orientation and position, the thinned version and boundary-detected versions being generated from the normalized second version image.

7. A system according to claim 6 wherein the thinned version is generated from the normalized second version image in a line thinning routine, the boundary-detected version is generated from the normalized second version image, the second version image being normalized with respect to writing line width and writing line size in the size normalization routine to provide the boundary detected image.

8. A system according to claim 2 wherein the segments comprise a first set of four quadrants defined from a center of gravity of the signature image, a second independent set of two vertical zones on either side of the center of gravity and a third independent set of two horizontal zones on either side of the center of gravity.

9. A system according to claim 2 wherein the extracted signature features are evaluated for their distance from a reference obtained from the evaluation program, the distance being computed from at least one genuine signature and being used to effect the authentication decision according to a threshold.

10. A system according to claim 9 wherein the threshold is adapted by the evaluation program according to new data from genuine signatures received by the evaluation program to enhance the decision.

11. A system according to claim 2 wherein two complex element signature features are extracted by matching and mismatching a horizontal profile of the signature image with a mean horizontal profile of a training sample comprising at least one genuine signature.

12. A system according to claim 2 wherein the evaluation program determines a feature set of said extracted signature features to be used by the signature processing program for the decision.

13. A system according to claim 2 wherein said authentic signature database includes image data from actual signatures on documents wherein said authentic signature data includes signatures of a person used to writing English, a person used to writing Japanese, a person used to writing Korean or a person used to writing Arabic.

14. A system according to claim 2, wherein said authentic signature database includes signature image data having a resolution of 80 dpi or less.

15. A system according to claim 2 wherein said authentic signature database includes at least one thousand individual records.

16. A system according to claim 2 wherein said authentication decision is qualified by a self evaluation routine to maintain a maximum ratio of an average of the percentage of correct acceptance decisions and an average of correct rejection.

17. A system according to claim 2 comprising a bank check signature verification system.

18. A system for automatic signature authentication comprising:

a) a computer-implementable signature processing program to examine a digitized signature image obtained from a document, optionally a check, the signature image purportedly being the signature of a specific person, to generate an authentication decision authenticating the signature as genuine or rejecting the signature as a forgery;

b) an authenticated signature database of signature features for use by signature processing program in generating authentication decision; and

c) an evaluation program to evaluate the results produced by the signature processing program and to provide update information to the signature database for use by signature processing program regarding examined signatures accepted as genuine by the signature processing program;

wherein.

d) the digitized signature image is scanned from a paper document;

e) multiple versions of the signature image are electronically generated and examined;

f) the results of examination are statistically compiled to enhance the authentication decision;

g) the multiple versions of the signature image comprise an unmodified version, a thinned version and a boundary-detected version;

h) the unmodified version is obtained by cleaning the signature image and is passed directly to a dissimilarity measurement procedure; and

i) a second version of the cleaned signature image is processed in a normalization procedure to adjust the image to a system standardized configuration with regard to orientation and position;

j) the thinned version is generated from the normalized second version image in a line thinning routine; and

k) the boundary-detected version is generated from the normalized second version image by normalization with respect to writing line width and writing line size.

19. A system for automatic signature authentication comprising:

a) a computer-implementable signature processing program to examine a digitized signature image obtained from a document, optionally a check, the signature image purportedly being the signature of a specific person, to generate an authentication decision authenticating the signature as genuine or rejecting the signature as a forgery;

b) an authenticated signature database of signature features for use by signature processing program in generating authentication decision; and

c) an evaluation program to evaluate the results produced by the signature processing program and to provide update information to the signature database for use by signature processing program regarding examined signatures accepted as genuine by the signature processing program;

wherein;

d) the digitized signature image is scanned from a paper document;

e) multiple versions of the signature image are electronically generated and examined;

f) the results of examination are statistically compiled to enhance the authentication decision;

g) each version of the signature image is defined into segments, signature features are extracted from individual segments and the extracted signature features are evaluated against training data received from the evaluation program; and

h) the segments comprise a first set of four quadrants defined from a center of gravity of the signature image, a second independent set of two vertical zones on either side of the center of gravity and a third independent set of two horizontal zones on either side of the center of gravity.

20. A system according to claim 19 wherein horizontal, vertical, positive and negative slant features in each said segment are computed and output as signature features to the signature database.

21. A system for automatic signature authentication comprising:

a) a computer-implementable signature processing program to examine a digitized. signature image obtained from a document, optionally a check, signature image purportedly being the signature of a specific person, to generate an authentication decision authenticating the signature as genuine or rejecting the signature as a forgery; and

b) an authenticated signature database of signature features for use by the signature processing program in generating the authentication decision;

wherein;

c) multiple versions of the signature image are electronically generated and examined, the examination results being statistically compiled to enhance the authentication decision;

d) the multiple versions of the signature image comprise an unmodified version, a thinned version and a boundary-detected version;

e) the unmodified version is obtained by cleaning the signature image and is passed directly to a dissimilarity measurement procedure;

f) a second version of the cleaned signature image is processed in a normalization procedure to adjust the image to a system standardized configuration with regard to orientation and position,

g) thinned and boundary-detected versions are electronically generated from the normalized second version image;

h) the thinned version is generated from the normalized second version image in a line thinning routine;

i) the boundary-detected version is generated from the normalized second version image;

j) the normalized second version image is normalized with respect to writing line width and writing line size in the size normalization routine to provide the boundary detected image; and

k) each version of the signature image is defined into segments, signature features are extracted from individual segments and extracted signature features are evaluated against training data received from an evaluation program.

22. A system according to claim 21 wherein the segments comprise a first set of four quadrants defined from a center of gravity of the signature image, a second independent set of two vertical zones on either side of the center of gravity and a third independent set of two horizontal zones on either side of the center of gravity.

23. A system according to claim 22 wherein horizontal, vertical, positive and negative slant features in each said segment are computed and output as signature features to the signature database and wherein the extracted signature features are each analyzed globally for the complete signature and locally for each segment.

24. A computer-implemented method of automatic signature authentication comprising:

a) employing a signature processing program to examine a digitized signature image obtained from a document, optionally a check, the signature image purportedly being the signature of a specific person;

b) employing an authenticated signature database of signature features to generate an authentication decision authenticating the examined signature as genuine or rejecting the signature as a forgery; and

c) employing an evaluation program to evaluate the results produced by the signature processing program and provide update information to the signature database for use by the signature processing program regarding examined signatures accepted as genuine by the signature processing program;

wherein the examined signature is evaluated as genuine, the database contains less than a desired number of genuine samples and the method further comprises;

d) adding the examined sample signature data to the database and recomputing reference statistics for use in generating the authentication decision using the added examined sample signature data wherein the reference statistics comprise a signature feature set and a decision threshold;

e) employing a global signature feature database including signature features of genuine and forged samples of signatures of persons other than the specific person'"'"'s signature together with available genuine samples of the specific person'"'"'s signature;

f) employing the global signature database for signature recognition until the database includes a desired number of genuine samples of the specific person'"'"'s signature;

g) defining the signature image into segments wherein the segments comprise a first set of four quadrants defined from a center of gravity of the signature image, a second independent set of two vertical zones on either side of the center of gravity and a third independent set of two horizontal zones on either side of the center of gravity;

h) extracting signature features from individual segments; and

i) extracting and evaluating signature features against training data received from the evaluation program.