On-line signature verification

• US 6,587,577 B1
• Filed: 04/21/1999
• Issued: 07/01/2003
• Est. Priority Date: 04/21/1999
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Finkelstein, Lev
• Primary Examiner(s): Mehta, Bhavesh M.
• Assistant Examiner(s): DESIRE, GREGORY M
• Application Number:
  US09/295,944
• Time in Patent Office: 1,532 Days
• Field of Search: 382/155-159, 382/116, 382/119, 382/123, 382/224, 382/187, 382/209, 382/160
• US Class Current: 382/123
• CPC Class Codes: G06V 40/30 Writer recognition; Reading and verifying signatures

Abstract

A method of determining whether a test sample belongs to the same individual as a plurality of reference samples. The method includes providing respective values of one or more identifying features from the test sample and the plurality of reference samples, representing the test and reference samples by respective points having coordinates in a multi-dimensional space indicative of the respective feature values, and deciding whether the test sample belongs to the same individual as the plurality of reference samples based on a geometrical property of the points.

Claims

1. A method of determining whether a test sample belongs to the same individual as a plurality of reference samples, comprising:

providing respective values of one or more identifying features from the test sample and the plurality of reference samples;

representing the test and reference samples by respective points having coordinates in a multi-dimensional space indicative of the respective feature values, so as to define a polyhedron in the multi-dimensional space; and

deciding whether the test sample belongs to the same individual as the plurality of reference samples based on a geometrical property of the polyhedron.

2. A method according to claim 1, wherein providing the respective values comprises extracting from the samples a plurality of values of the one or more features descriptive of the samples, and wherein representing the samples by one or more geometric points comprises representing each sample by one point for each of the one or more features.

3. A method according to claim 2, wherein at least some of the one or more features comprise a value dependent on a comparison of two of the samples.

4. A method according to claim 2, wherein representing each of the test and reference samples by one or more geometric points comprises generating for each of the one or more features, a square matrix in which each row and each column is representative of one of the samples, and in which the values of the terms of the matrix are responsive to the value of the feature of the samples of the row and column of the term, and generating for each matrix a plurality of points representative of the matrix.

5. A method according to claim 4, wherein generating the plurality of points representative of the matrix comprises smoothing the matrix and generating the points based on the smoothed matrix.

6. A method according to claim 5, wherein smoothing the matrix comprises applying a square root operator to the terms of the matrix.

7. A method according to claim 5, wherein smoothing the matrix comprises generating a symmetrical matrix responsive to the matrix to be smoothed.

8. A method according to claim 1, wherein representing the test and reference samples by respective points comprises generating the points such that the distances between the points are proportional to the respective values of the features.

9. A method according to claim 1, wherein deciding based on the geometrical property of the polyhedron comprises deriving a parameter indicative of the regularity of the polyhedron.

10. A method according to claim 9, wherein deriving the parameter comprises deriving parameters indicative of the relative form of a polyhedron including points representing both the test and reference samples compared to a polyhedron including points representing only the reference samples.

11. A method according to claim 10, wherein deriving parameters indicative of the relative form of the polyhedrons comprises deriving relative heights and variances of the polyhedrons.

12. A method according to claim 10, wherein deciding whether the test sample belongs to the same individual as the reference samples comprises inputting the parameters to a classifier.

13. A method according to claim 12, wherein the classifier comprises a neural network.

14. A method according to claim 1, wherein the samples comprise signatures.

15. A method according to claim 14, wherein providing the respective values comprises determining a speed mismatch of the samples.

16. A method according to claim 1, wherein providing the respective values comprises performing a dynamic time warping technique on the samples.

17. A method according to claim 1, wherein representing the test and reference samples by the respective points comprises representing the samples by n-dimensional points wherein n is the number of reference samples.

18. A method according to claim 17, wherein representing the test and reference samples by the respective points comprises representing the samples such that each of the points has a different number of non-zero coordinates.

19. A method according to claim 18, wherein the test sample is represented by a point substantially all of whose coordinates are non-zero.

20. A method according to claim 1, and comprising eliminating one or more of the reference samples prior to representing the test and reference samples by the respective points.

21. Apparatus for determining whether a test sample belongs to the same individual as a plurality of reference samples, comprising:

an input device which acquires the samples;

a processor which determines values of one or more features of the samples, represents the samples by corresponding points in a multi-dimensional space having coordinates indicative of the feature values, so as to define a polyhedron in the multi-dimensional space, and calculates a geometrical property of the polyhedron; and

a classifier which determines whether the test sample belongs to the same individual as the plurality of reference samples responsive to the geometrical property.

22. Apparatus according to claim 21, wherein the samples comprise signatures, and the input device comprises a writing implement.

23. Apparatus according to claim 22, wherein the input device does not determine a pressure of the writing implement on a pad.

24. Apparatus according to claim 21, wherein the classifier comprises a neural network.

25. Apparatus according to claim 21, wherein the classifier is implemented by software in the processor.

26. Apparatus according to claim 21, wherein the processor represents each sample by one point for each of the one or more features.

27. Apparatus according to claim 21, wherein the processor determines values of the one or more features using a dynamic time warping technique.

28. Apparatus according to claim 21, wherein the processor calculates the geometrical property by deriving parameters indicative of the relative form of a polyhedron including points representing both the test and reference samples compared to a polyhedron including points representing only the reference samples.

29. A computer program product having computer readable program code embodied therein, which code causes a processor receiving a test sample and a plurality of reference samples, wherein each sample is characterized by respective values of one or more identifying features, to represent the test and reference samples by respective points having coordinates in a multi-dimensional space indicative of the respective feature values, so as to define a polyhedron in the multi-dimensional space, and to decide whether the test sample belongs to the same individual as the plurality of reference samples based on a geometrical property of the polyhedron.