Methods and systems for image fingerprinting

• US 8,406,456 B2
• Filed: 11/20/2008
• Issued: 03/26/2013
• Est. Priority Date: 11/20/2008
• Status: Active Grant

• Current Assignee: Freedom Solutions Group, LLC
• Original Assignee: Workshare Technology Incorporated
• Inventors: More, Scott
• Primary Examiner(s): Allison, Andrae S
• Application Number:
  US12/275,185
• Publication Number:
  US 20100124354A1
• Time in Patent Office: 1,587 Days
• Field of Search: 382/100, 382/101, 382/176, 382/190, 382/286, 705/58, 705/406, 380/55, 380/202, 713/176
• US Class Current: 382/100
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

The present invention provides methods and systems to protect an organization'"'"'s secure image information from unauthorized disclosure. In one embodiment, methods and systems to generate image fingerprints are provided. The fingerprints are generated for each feature point of an image data. Because the fingerprints take into account the neighborhood features around each feature point, the image fingerprints are robust against derivate images where the original image may have been altered. Methods and systems to maintain a fingerprint database for an organization'"'"'s secure image data is also provided. In one embodiment, client fingerprints are generated for image data that a user intends to transmit outside of the organization. In some embodiments, suitable security actions are initiated if any of the client fingerprints match any of the fingerprints in the fingerprint database.

Claims

1. A computer implemented method for fingerprinting image data, the computer implemented method comprising:

selecting a first feature point in the image data;

identifying a plurality of neighborhood feature points closest in distance to the first feature point, wherein the plurality of neighborhood feature points closest in distance to the first feature point includes a P-neighborhood, the P-neighborhood including P number of neighborhood feature points closest in distance to the first feature point and feature points in the P-neighborhood are determined by storing all feature points of the image data in a feature point table, and determining P number of feature points closest to the location of the first feature point within the feature point table;

generating a plurality of point vectors, each point vector computed based on distance and angle between a particular neighborhood feature point and the first feature point; and

aggregating the plurality of point vectors to generate a fingerprint corresponding to the first feature point.

2. A computer implemented method for fingerprinting image data as recited in claim 1, the computer implemented method further comprising:

converting the image data to a canonical format.

3. A computer implemented method for fingerprinting image data as recited in claim 2, wherein the canonical format includes mapping the image data to a two dimensional array defining intensity values of the image data at each pixel location.

4. A computer implemented method for fingerprinting image data as recited in claim 1, wherein feature points in the P-neighborhood are determined by comparing the first feature point with all other feature points of the image data.

5. A computer implemented method for fingerprinting image data as recited in claim 1, wherein distance between the first feature point and another feature point of the image data is computed using a Euclidian distance formula.

6. A computer implemented method for fingerprinting image data as recited in claim 5, wherein a final square-root function is not computed in the Euclidian distance formula.

7. A computer implemented method for fingerprinting image data as recited in claim 1, wherein the fingerprint corresponding to the first feature point is generated by applying a hash function over the plurality of point vectors.

8. A computer implemented method for fingerprinting image data as recited in claim 7, wherein the fingerprint corresponding to the first feature point includes an integer value.

9. A computer implemented method for fingerprinting image data, the computer implemented method comprising:

selecting a first feature point in the image data;

identifying a plurality of neighborhood feature points closest in distance to the first feature point, wherein the plurality of neighborhood feature points closest in distance to the first feature point includes a P-neighborhood, the P-neighborhood including P number of neighborhood feature points closest in distance to the first feature point;

generating a plurality of point vectors, each point vector computed based on distance and angle between a particular neighborhood feature point and the first feature point; and

aggregating the plurality of point vectors to generate a fingerprint corresponding to the first feature point; and

designating as an anchor point a least distant feature point, the least distant feature point being a specific feature point closest to the first feature point within the P-neighborhood.

10. A computer implemented method for fingerprinting image data as recited in claim 9, wherein the distance and angle between the first feature point and the particular neighborhood feature point is computed relative to the anchor point.

11. A computer implemented method for fingerprinting image data as recited in claim 10, wherein the distance between the particular neighborhood feature point and the first feature point is normalized relative to the distance between the anchor point and the first feature point.

12. A computer implemented method for fingerprinting image data as recited in claim 11, wherein the distance between the first feature point and the anchor point is normalized to a unitary value.

13. A computer implemented method for fingerprinting image data as recited in claim 10, wherein the angle between the particular neighborhood feature point and the first feature point is normalized relative to the angle between the anchor point and the first feature point.

14. A computer implemented method for fingerprinting image data as recited in claim 13, wherein the angle is computed based on a predetermined direction of rotation.

15. A computer implemented method for fingerprinting image data as recited in claim 14, wherein the direction of rotation is determined for a specific P neighborhood based on a feature local to the specific P neighborhood.

16. A computer implemented method for fingerprinting image data as recited in claim 13, wherein the angle between the first feature point and the anchor point is normalized to a zero value.

17. A computer implemented method for fingerprinting image data as recited in claim 10, wherein the values of the normalized distance and angle between the particular neighborhood feature point and the first feature point are rounded to enable generation of substantially similar fingerprints across related features of the image data.

18. A computer implemented method for fingerprinting image data, the computer implemented method comprising:

selecting a first feature point in the image data;

selecting a plurality of neighborhood feature points closest in distance to the first feature point, wherein a least distant feature point, measured relative to the first feature point, of the plurality of neighborhood feature points is designated as an anchor point;

generating a plurality of point vectors, each point vector computed based on distance and angle between a particular neighborhood feature point and the first feature point, wherein the distance and angle are computed relative to the anchor point; and

aggregating the plurality of point vectors to generate a fingerprint corresponding to the first feature point.

19. A computer implemented method for fingerprinting image data as recited in claim 18, the computer implemented method further comprising:

converting the image data to a canonical format.

20. A computer implemented method for fingerprinting image data as recited in claim 19, wherein the canonical format includes mapping the image data to a two dimensional array defining intensity values of the image data at each pixel location.

21. A computer implemented method for fingerprinting image data as recited in claim 18, wherein the plurality of neighborhood feature points closest in distance to the first feature point includes a P-neighborhood, the P-neighborhood including P number of neighborhood feature points closest in distance to the first feature point.

22. A computer implemented method for fingerprinting image data as recited in claim 21, wherein feature points in the P-neighborhood are determined by comparing the first feature point with all other feature points of the image data.

23. A computer implemented method for fingerprinting image data as recited in claim 21, wherein feature points in the P-neighborhood are determined by storing all feature points of the image data in a feature point table, and determining P number of feature points closest to the location of the first feature point within the feature point table.

24. A computer implemented method for fingerprinting image data as recited in claim 18, wherein the distance between the particular neighborhood feature point and the first feature point is normalized relative to the distance between the anchor point and the first feature point.

25. A computer implemented method for fingerprinting image data as recited in claim 24, wherein the distance between the first feature point and the anchor point is normalized to a unitary value.

26. A computer implemented method for fingerprinting image data as recited in claim 18, wherein the angle between the particular neighborhood feature point and the first feature point is normalized relative to the angle between the anchor point and the first feature point.

27. A computer implemented method for fingerprinting image data as recited in claim 26, wherein the angle between the first feature point and the anchor point is normalized to a zero value.

28. A computer implemented system for fingerprinting image data, the computer implemented system comprising:

a processor to execute the operations of an image fingerprinting module, the image fingerprinting module including;

a selecting means for selecting a first feature point in the image data and for identifying a plurality of neighborhood feature points closest in distance to the first feature point;

a vector computing means for generating a plurality of point vectors, each point vector computed based on distance and angle between a particular neighborhood feature point and the first feature point; and

a hash generating means for aggregating the plurality of point vectors and generating a fingerprint corresponding to the first feature point; and

a memory coupled with the processor, the memory storing instructions corresponding to the operations of the image fingerprinting module;

wherein the plurality of neighborhood feature points closest in distance to the first feature point includes a P-neighborhood, the P-neighborhood including P number of neighborhood feature points closest in distance to the first feature point; and

wherein the selecting means determines feature points in the P-neighborhood by storing all feature points of the image data in a feature point table, and then determining P number of feature points closest to the location of the first feature point within the feature point table.

29. A computer implemented system for fingerprinting image data as recited in claim 28, wherein the fingerprinting module further includes:

an image converting means for converting the image data to a canonical format.

30. A computer implemented system for fingerprinting image data as recited in claim 29, wherein the canonical format includes mapping the image data to a two dimensional array defining intensity values of the image data at each pixel location.

31. A computer implemented system for fingerprinting image data as recited in claim 28, wherein the selecting means determines feature points in the P-neighborhood by comparing the first feature point with all other feature points of the image data.

32. A computer implemented system for fingerprinting image data as recited in claim 28, wherein the selecting means computes distance between the first feature point and an other feature point of the image data using a Euclidian distance formula.

33. A computer implemented system for fingerprinting image data as recited in claim 32, wherein a final square-root function is not computed in the Euclidian distance formula.

34. A computer implemented system for fingerprinting image data as recited in claim 28, wherein the hash generating means generates the fingerprint corresponding to the first feature point by applying a hash function over the plurality of point vectors.

35. A computer implemented system for fingerprinting image data as recited in claim 34, wherein the fingerprint corresponding to the first feature point includes an integer value.

36. A computer implemented system for fingerprinting image data, the computer implemented system comprising:

a processor to execute the operations of an image fingerprinting module, the image fingerprinting module including;

a selecting means for selecting a first feature point in the image data and for identifying a plurality of neighborhood feature points closest in distance to the first feature point;

a vector computing means for generating a plurality of point vectors, each point vector computed based on distance and angle between a particular neighborhood feature point and the first feature point; and

a hash generating means for aggregating the plurality of point vectors and generating a fingerprint corresponding to the first feature point;

a memory coupled with the processor, the memory storing instructions corresponding to the operations of the image fingerprinting module; and

wherein the plurality of neighborhood feature points closest in distance to the first feature point includes a P-neighborhood, the P-neighborhood including P number of neighborhood feature points closest in distance to the first feature point, and wherein the vector computing means designates, relative to the first feature point, a least distant feature point within the P-neighborhood as an anchor point.

37. A computer implemented system for fingerprinting image data as recited in claim 36, wherein the vector computing means computes the distance and angle between the first feature point and the particular neighborhood feature point and the first feature point relative to the anchor point.

38. A computer implemented system for fingerprinting image data as recited in claim 37, wherein the vector computing means normalizes the distance between the particular neighborhood feature point and the first feature point relative to the distance between the anchor point and the first feature point.

39. A computer implemented system for fingerprinting image data as recited in claim 38, wherein the vector computing means normalizes the distance between the first feature point and the anchor point to a unitary value.

40. A computer implemented system for fingerprinting image data as recited in claim 37, wherein the vector computing means normalizes the angle between the particular neighborhood feature point and the first feature point relative to the angle between the anchor point and the first feature point.

41. A computer implemented method for fingerprinting image data as recited in claim 40, wherein the angle is computed based on a predetermined direction of rotation.

42. A computer implemented method for fingerprinting image data as recited in claim 41, wherein the direction of rotation is determined for a specific P neighborhood is based on a feature local to the specific P neighborhood.

43. A computer implemented system for fingerprinting image data as recited in claim 40, wherein the vector computing means normalizes the angle between the first feature point and the anchor point to a zero value.

44. A computer implemented system for fingerprinting image data as recited in claim 37, wherein the vector computing means rounds values of the normalized distance and angle between the particular neighborhood feature point and the first feature point to enable generation of substantially similar fingerprints across related features of the image data.

45. A computer implemented method for fingerprinting image data, the computer implemented method comprising:

selecting a first feature point in the image data;

identifying a plurality of neighborhood feature points closest in distance to the first feature point, wherein the plurality of neighborhood feature points closest in distance to the first feature point includes a P-neighborhood, the P-neighborhood including P number of neighborhood feature points closest in distance to the first feature point;

generating a plurality of point vectors, each point vector computed based on distance and angle between a particular neighborhood feature point and the first feature point;

aggregating the plurality of point vectors to generate a fingerprint corresponding to the first feature point;

wherein distance between the first feature point and an other feature point of the image data is computed using a Euclidian distance formula and a final square-root function is not computed in the Euclidian distance formula.