Image recognition

• US 7,689,043 B2
• Filed: 10/11/2004
• Issued: 03/30/2010
• Est. Priority Date: 10/09/2003
• Status: Active Grant

• Current Assignee: University of York
• Original Assignee: University of York
• Inventors: Heseltine, Thomas, Pears, Nicholas, Austin, James
• Primary Examiner(s): Chawan; Sheela C
• Application Number:
  US10/575,739
• Publication Number:
  US 20070122007A1
• Time in Patent Office: 1,996 Days
• Field of Search: 382/118, 382/190, 382/195, 382/103, 382/104, 382/170, 382/206, 382/155, 382/159, 382/115, 382/225, 382/117, 382/201, 382/224, 382/276, 704/251, 704/E15.042, 704/521, 348/77, 348/169, 385/116, 434/43, 345/419, 345/420, 345/49, 345/679
• US Class Current: 382/224
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions

Abstract

A camera A produces a 3D natural image B which is re-oriented and repositioned at N to predetermined parameters. Different processes C, D, E extract different features from the image to provide different processed images. The data space occupied by the processed images is reduced, for example by Principle Component Analysis, at F, G, H and the reduced processed images are combined at O to provide an image key I representative of the image B. The image key I is compared at J with stored image keys L of known images, and the output comparison is sorted at K to produce a final list M of potential matches. In a verification process, just a single image key L may be compared. The processed images at C, D, E may alternatively be combined prior to a single subspace reduction and/or optimisation method. 2D data may be combined with or used instead of the 3D data.

Claims

1. A method of recognizing an image, comprising the steps of utilizing a computer or processor to:

a. process the image to provide an image set containing a plurality of different processed images which are derived from the original image to be recognized;

b. combine the processed images in the image set;

c. transform the data space occupied by the processed images in the image set;

d. generate, from the image-set represented in the transformed data space, an image key representative of the image; and

e. compare the image key with at least one previously stored image key of a known image, to verify the identity of the image.

2. A method according to claim 1, wherein step a. includes extracting image features including at least one of edges, lines, wavelets, gradient components, curvature components and colour components.

3. A method according to claim 1, wherein step b. is carried out prior to step c.

4. A method according to claim 1, wherein step c. is carried out prior to step b.

5. A method according to claim 1, wherein step e. comprises comparing the image key with just one previously stored image key.

6. A method according to claim 1, wherein step e. comprises comparing the image key wit a plurality of previously stored image keys, to identify the image.

7. A method according to claim 6, comprising the further step of sorting the results of the comparison in step e. to produce a list of potential matches with previously stored image keys.

8. A method according to claim 6, wherein step e. is carried out using a Euclidean distance metric (the L2 norm), mahalanobis distance metric or a cosine distance metric.

9. A method according to claim 1, including the step prior to step a. of rotating and/or positioning the image to a predetermined orientation and/or position and/or depth normalisation.

10. A method according to claim 1, including a step prior to step b. of normalising data prior to combination.

11. A method according to claim 1, wherein said image is obtained from a camera.

12. A method according to claim 1, wherein said image comprises 3D data.

13. A method according to claim 12, wherein said image comprises a registered 2D-3D image pair.

14. A method according to claim 1, wherein said image comprises 2D data.

15. A method according to claim 1, wherein step c. is carried out by a Principal Component Analysis method.

16. A method according to any of claims 1 to 13, wherein step is carried out by Fisher'"'"'s Linear Discriminant Analysis method.

17. A method according to claim 1, wherein said image is an image of a face.

18. A method according to claim 1, wherein said image is an image of a human face.

19. A method according to claim 1, wherein said image is a natural image.

20. A method according to claim 1, wherein said image set includes the original image.

21. Apparatus for recognizing an image, the apparatus comprising:

a. processing means arranged to process the image to provide a plurality of different processed images which are derived from the original image to be recognized;

b. combining means arranged to combine the processed images;

c. reducing means arranged to reduce the data space occupied by the processed images;

d. generating means arranged to generate from the combined and reduced processed images an image key representative of the image; and

e. comparison means arranged to compare the image key with at least one previously stored image key of a known image, to verify the identity of the image.

22. Apparatus according to claim 21, wherein said processing means is arranged to extract image features including at least one of edges, lines, wavelets, gradient components, curvature components and colour components.

23. Apparatus according to claim 21, wherein said reducing means is arranged to reduce said data space after said combining means combines said processed images.

24. Apparatus according to claim 21, wherein said reducing means is arranged to reduce said data space before said combining means combines said processed images.

25. Apparatus according to claim 21, wherein said comparison means is arranged to compare the image key with just one previously stored image key.

26. Apparatus according to claim 21, wherein said comparison means is arranged to compare the image key with a plurality of previously stored image keys, to identify the image.

27. Apparatus according to claim 26, further comprising sorting means arranged to sort results from said comparison means to produce a list of potential matches with previously stored image keys.

28. Apparatus according to claim 26, wherein said comparison means is arranged to carry out a comparison using a Euclidean distance metric (the L2 norm), mahalanobis distance metric or a cosine distance metric.

29. Apparatus according to claim 21, further comprising normalising means arranged to normalise data prior to combination of the data by said combination means.

30. Apparatus according to claim 21, further comprising a camera that is arranged to provide said image to be recognised.

31. Apparatus according to claim 21, arranged to process 3D image data.

32. Apparatus according to claim 21, arranged to process 2D image data.

33. Apparatus according to claim 21, arranged to process an image that comprises a registered 2D-3D image pair.

34. Apparatus according to claim 21, wherein said reducing means is arrange to carry out a Principal Component Analysis method.

35. Apparatus according to claim 21, wherein said reducing means is arrange to carry out a Fisher'"'"'s Linear Discriminant Analysis method.

36. Apparatus according to claim 21, arranged to process an image of a face.

37. Apparatus according to claim 21, arranged to process an image of a human face.

38. Apparatus according to claim 21, arranged to process a natural image.

39. Apparatus according to claim 21, wherein one of said processed images includes the original image.

40. A method of recognizing an image, comprising the steps of utilizing a computer or processor to:

a. transform the data space occupied by a three dimensional image using Fisher'"'"'s Linear Discriminant Analysis;

b. generate, from the transformed data space, an image key representative of the image; and

c. compare the image key with at least one previously stored image key of a known image, to verify the identity of the image.

41. Apparatus for recognizing a three-dimensional image, the apparatus comprising:

a. means for transforming the data space occupied by a three dimensional image using Fisher'"'"'s Linear Discriminant Analysis;

b. means for generating, from the transformed data space, an image key representative of the image; and

c. means for comparing the image key with at least one previously stored image key of a known image, to verify the identity of the image.