Signature recognition system and method

• US 6,661,908 B1
• Filed: 01/13/2000
• Issued: 12/09/2003
• Est. Priority Date: 01/13/1999
• Status: Expired due to Term

• Current Assignee: Computer Associates Think Inc. (Broadcom, Inc.)
• Original Assignee: Computer Associates Think Inc. (Broadcom, Inc.)
• Inventors: Avnl, Yossi, Suchard, Eytan
• Primary Examiner(s): Mariam, Daniel G.
• Application Number:
  US09/482,075
• Time in Patent Office: 1,426 Days
• Field of Search: 382/115-116, 382/119, 382/121, 382/159, 382/181, 382/187-190, 382/197, 382/202, 382/209, 382/217-218, 382/224-225, 382/314, 706/20, 706/25, 706/31, 713/201, 178/18.01, 178/18.03, 178/19.01, 345/173, 345/179, 348/161
• US Class Current: 382/119
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06F 18/24137 Distances to cluster centroïds
• CPC Class Codes: G06Q 40/08 Insurance
• CPC Class Codes: G06V 40/30 Writer recognition; Reading and verifying signatures

Abstract

A method of authenticating a signature including the steps of sampling a signature and storing data representative of the signature, converting the data to high dimensions vectors, feeding the high dimension vectors to an unsupervised neural network, performing a high order principal component extraction process on the high dimensions vectors to thereby identifying clusters of high dimension points, and analyzing the clusters of high dimension points to determine, based on previously stored information, the authenticity of the signature. Also an apparatus for such authentication including a sampling device for sampling a signature and storing data representative of the signature, a converting device connected downsteam of the sampling device for converting the data to high dimension vectors, an unsupervised neural network for receiving the high dimension and performing a high order principal component extraction process on the high dimensions vectors to thereby identify clusters of high dimension points, and an analyzing device connected to the unsupervised neural network for analyzing the clusters of high dimension points to determine the authenticity of the signature.

Claims

1. A method of authenticating a signature, comprising the steps of:

sampling a signature and storing data representative of said signature;

converting said data to high dimension vectors using a recursive sampling process, wherein said recursive sampling process comprises iteratively focusing on increasingly smaller features of said signature, defined by shorter time spans;

feeding said high dimension vectors to an unsupervised neural network and performing a high order principal component extraction process on said high dimension vectors by cumulative orthonormalization, thereby identifying clusters of high dimension points; and

analyzing said clusters of high dimension points to determine, based on previously stored information, the authenticity of said signature.

2. The method of claim 1, further comprising the step of digitally sampling said signature.

3. The method of claim 1, wherein said sampling is effected via at least one of a mouse, a pressure sensitive pad, a digital tablet, a stylus and an electronic pen.

4. The method of claim 3, further comprising sampling pressure applied to said at least one of said mouse, pressure sensitive pad, digital tablet, stylus and electronic pen.

5. The method of claim 1, further comprising generating a list of time/location pairs.

6. The method of claim 1, wherein sampling is effected at a rate of at least about 40 samples per second.

7. The method of claim 1, wherein said recursive sampling process comprises 12 iterations.

8. The method of claim 1, wherein a time period associated with each iteration is in accordance with the following criteria:

(a) the time period examined in the first iteration is between 50% and 70% of a signature time (with steps of 2%);

(b) the time period examined in a second to 13th iterations is 70% of the time period examined in the previous iteration; and

(c) the time period in each iteration is either in the start of the time period of the previous iteration, or in its end, but not in the middle.

9. The method of claim 1, wherein said recursive process lasts no more than 3 seconds.

10. The method of claim 1, wherein a neuron is prevented from growing to an extent whereby that neuron recognizes all vectors as a belonging to a single cluster.

11. The method of claim 1, wherein the clusters are circular or bubble-shaped.

12. The method of claim 11, wherein a radius of a bubble corresponds to the standard deviation of the distance of the vectors for a winning neuron.

13. A method of authenticating a signature, comprising the steps of:

sampling a signature and storing data representative of said signature;

converting said data to high dimension vectors using a recursive sampling process;

feeding said high dimension vectors to an unsupervised neural network and performing a high order principal component extraction process on said high dimension vectors by cumulative orthonormalization, thereby identifying clusters of high dimension points;

analyzing said clusters of high dimension points to determine, based on previously stored information, the authenticity of said signature;

generating said information by;

providing a plurality of sample signatures;

effecting said sampling, converting and feeding steps for each of said signatures;

computing a temporal summation r and an average temporal summation s for each of said signatures; and

based on said computed r and s values, outputting a measure of global signature structure deviation A and a measure of local signature structure deviation B.

14. The method of claim 13, wherein r is a ratio of the number of vectors within an ellipsoid to the total number of vectors and s is the average of distances of all vectors within said ellipsoid.

15. The method of claim 13, wherein said previously stored information is compared with said signature to determine, based on a predetermined criteria, the authenticity thereof.

16. The method of claim 13, further comprising multiplying values A and B including and determining whether the product is less than 1.

17. The method of claim 16, further comprising executing said multiplying step in a Pi neuron.

18. The method of claim 17, wherein 20 to 40 Pi neurons are employed.

19. A method of authenticating a signature, comprising the steps of:

sampling a signature and storing data representative of said signature;

converting said data to high dimension vectors using a recursive sampling process;

feeding said high dimension vectors to an unsupervised neural network and performing a high order principal component extraction process on said high dimension vectors by cumulative orthonormalization, thereby identifying clusters of high dimension points;

analyzing said clusters of high dimension points to determine, based on previously stored information, the authenticity of said signature; and

assessing the presence of overgeneralization in said neural network, wherein a condition of overgeneralization is deemed to occur when at least one of a ratio of the number of vectors within an ellipsoid to the total number of vectors squared (r²) and an average of distances of all vectors within said ellipsoid squared (S²) divided by the variance of the value r or s, respectively, is greater than a predetermined threshold.

20. Software stored in a computer storage medium for authenticating a signature, the software operable to:

sample a signature and storing data representative of the signature;

convert the data to high dimension vectors using a recursive sampling process, the recursive sampling process comprising iteratively focusing on increasingly smaller features of the signature, defined by shorter time spans;

feed the high dimension vectors to an unsupervised neural network and perform a high order principal component extraction process on the high dimension vectors by cumulative orthonormalization, thereby identifying clusters of high dimension points; and

analyze the clusters of high dimension points to determine, based on previously stored information, the authenticity of the signature.

21. The software of claim 20, further comprising generating a list of time/location pairs.

22. The software of claim 20, wherein the clusters are circular or bubble-shaped.

23. A system for authenticating a signature comprising:

at least one memory operable to store data representatives of a plurality signatures; and

one or more processors, collectively operable to;

sample a first signature and storing a data representative of the first signature in memory;

convert the data to high dimension vectors using a recursive sampling process, the recursive sampling process comprising iteratively focusing on increasingly smaller features of said signature, defined by shorter time spans;

feed the high dimension vectors to an unsupervised neural network and perform a high order principal component extraction process on the high dimension vectors by cumulative orthonormalization, thereby identifying clusters of high dimension points; and

analyze the clusters of high dimension points to determine, based on previously stored information, the authenticity of the first signature.

24. The system of claim 23, the one or more processors further operable to generate a list of time/location pairs.

25. The system of claim 23, wherein the clusters are circular or bubble-shaped.

26. A system for authenticating a signature, comprising:

means for sampling a signature and storing data representative of the signature;

means for converting the data to high dimension vectors using a recursive sampling process, the recursive sampling process comprises iteratively focusing on increasingly smaller features of the signature, defined by shorter time spans;

means for feeding the high dimension vectors to an unsupervised neural network and performing a high order principal component extraction process on the high dimension vectors by cumulative orthonormalization, thereby identifying clusters of high dimension points; and

means for analyzing the clusters of high dimension points to determine, based on previously stored information, the authenticity of the signature.