Method and system of person identification by facial image

• US 8,184,914 B2
• Filed: 10/10/2011
• Issued: 05/22/2012
• Est. Priority Date: 11/29/2007
• Status: Active Grant

• Current Assignee: Viewdle, Inc.
• Original Assignee: Viewdle, Inc.
• Inventors: Kyyko, Volodymyr, Matsello, Vyacheslav, Tsariov, Andriy, Musatenko, Yuriy, Schlesinger, Mykhailo, Kyiko, Kostiantyn, Nedashkivskyi, Oleksii
• Primary Examiner(s): MARIAM, DANIEL G
• Application Number:
  US13/270,155
• Publication Number:
  US 20120027270A1
• Time in Patent Office: 225 Days
• Field of Search: 382/118, 382/173, 382/209, 382/218, 382/250, 382/254, 382/276
• US Class Current: 382/218
• CPC Class Codes: G06V 20/647 by matching two-dimensional images to three-dimensional objects
• CPC Class Codes: G06V 40/165 using facial parts and geometric relationships
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

A method to recognize a facial image is described. An input facial image is normalized by scaling and rotation angle using methods of eye pupil centers detection. The input facial image is further normalized by lighting intensity. Template images are obtained either by the processing of certain images taken from different face positions, or by a preliminary reconstruction of a 3D face model based on stereo-pair images. Using the 3D model template, facial images are generated at different rotation angles. Distances between the input facial image and the template image are calculated from the Discrete Cosine Transformation (DCT) features, defined by overlapped blocks of these images. The facial image is recognized based on these distances.

Claims

1. A computer-based method of comparing an image with a template image, the method comprising:

deriving a normalized image based on the image;

splitting the normalized image into rectangular blocks;

representing the set of rectangular blocks of the normalized image by a DCT (discrete cosine transform);

calculating a first distance between the normalized image and the template image based on a weighted sum of distances between corresponding blocks from the set of rectangular blocks of the normalized image and a set of rectangular blocks of the template image;

comparing the first distance to a first predetermined threshold, the first predetermined threshold indicating a degree of similarity between the normalized image and the template;

responsive to the first distance being greater than the first predetermined threshold, determining a set of overlapping blocks from the normalized image and representing the set of overlapping blocks by the DCT;

calculating a second distance between the normalized image and the template image based on a weighted sum of distances between corresponding blocks from the set of overlapping blocks of the normalized image a set of overlapping blocks of the template image; and

recognizing the normalized image based on the calculated distances.

2. The method of claim 1, wherein deriving the normalized image based on the image comprises adjusting lighting conditions of the image by one or both of the following:

a histogram equalization method and a local contrast calculation method.

3. The method of claim 1, wherein representing the rectangular blocks by a DCT comprises, for each rectangular block:

calculating a table of DCT coefficients;

extracting the most informative of the DCT coefficients via a zigzag scan of the DCT coefficients of the block starting from a coefficient located at the top-left corner of the table of the DCT coefficients;

replacing one or more of the DCT coefficients by weighted differences of corresponding coefficients located in neighboring horizontal and vertical blocks; and

obtaining resultant feature values for the block.

4. The method of claim 1, wherein calculating the distances between the normalized image and the template comprises determining a sum of absolute values of the distances between corresponding features of the normalized image and the existing template divided by the sum of the absolute values of the corresponding features.

5. The method of claim 1, wherein corresponding blocks have the same coordinates.

6. The method of claim 1, wherein recognizing the normalized image based on the second distance calculated results in a rejection of one or more other templates when a first distance calculated between the normalized image and another template is above a second predetermined threshold.

7. The method of claim 1, wherein the normalized image is associated with an unknown object and the template image is associated with a first known object and recognizing the normalized image comprises determining a degree of similarity between the normalized image and the template and degrees of similarity between the normalized image and templates associated with any other known object, a degree of similarity being based on a distance between the normalized image and a template.

8. The method of claim 7, wherein the objects are human faces.

9. A non-transitory machine-readable storage medium storing executable computer program instructions for comparing an image with a template image, the instructions performing steps comprising:

deriving a normalized image based on the image;

splitting the normalized image into rectangular blocks;

representing the set of rectangular blocks of the normalized image by a DCT (discrete cosine transform);

calculating a first distance between the normalized image and the template image based on a weighted sum of distances between corresponding blocks from the set of rectangular blocks of the normalized image and a set of rectangular blocks of the template image;

comparing the first distance to a first predetermined threshold, the first predetermined threshold indicating a degree of similarity between the normalized image and the template;

responsive to the first distance being greater than the first predetermined threshold, determining a set of overlapping blocks from the normalized image and representing the set of overlapping blocks by the DCT;

calculating a second distance between the normalized image and the template image based on a weighted sum of distances between corresponding blocks from the set of overlapping blocks of the normalized image a set of overlapping blocks of the template image; and

recognizing the normalized image based on the calculated distances.

10. The computer readable medium of claim 9, wherein deriving the normalized image based on the image comprises adjusting lighting conditions of the image by one or both of the following:

a histogram equalization method and a local contrast calculation method.

11. The computer readable medium of claim 9, wherein representing the rectangular blocks by a DCT comprises, for each rectangular block:

calculating a table of DCT coefficients;

extracting the most informative of the DCT coefficients via a zigzag scan of the DCT coefficients of the block starting from a coefficient located at the top-left corner of the table of the DCT coefficients;

replacing one or more of the DCT coefficients by weighted differences of corresponding coefficients located in neighboring horizontal and vertical blocks; and

obtaining resultant feature values for the block.

12. The computer readable medium of claim 9, wherein calculating the distances between the normalized image and the template comprises determining a sum of absolute values of the distances between corresponding features of the normalized image and the existing template divided by the sum of the absolute values of the corresponding features.

13. The computer readable medium of claim 9, wherein corresponding blocks have the same coordinates.

14. The computer readable medium of claim 9, wherein recognizing the normalized image based on the second distance calculated results in a rejection of one or more other templates when a first distance calculated between the normalized image and another template is above a second predetermined threshold.

15. The computer readable medium of claim 9, wherein the normalized image is associated with an unknown object and the template image is associated with a first known object and recognizing the normalized image comprises determining a degree of similarity between the normalized image and the template and degrees of similarity between the normalized image and templates associated with any other known object, a degree of similarity being based on a distance between the normalized image and a template.

16. The method of claim 15, wherein the objects are human faces.

17. A computer-based system for comparing an image with a template image, the system comprising:

a non-transitory computer-readable storage medium comprising executable computer program instructions that when executed by one or more processors cause the processors to perform one or more steps, the steps causing the processors to;

derive a normalized image based on the image;

split the normalized image into rectangular blocks;

represent the set of rectangular blocks of the normalized image by a DCT (discrete cosine transform);

calculate a first distance between the normalized image and the template image based on a weighted sum of distances between corresponding blocks from the set of rectangular blocks of the normalized image and a set of rectangular blocks of the template image;

compare the first distance to a first predetermined threshold, the first predetermined threshold indicating a degree of similarity between the normalized image and the template;

in response to the first distance being greater than the first predetermined threshold, determine a set of overlapping blocks from the normalized image and represent the set of overlapping blocks by the DCT;

calculate a second distance between the normalized image and the template image based on a weighted sum of distances between corresponding blocks from the set of overlapping blocks of the normalized image a set of overlapping blocks of the template image; and

recognize the normalized image based on the calculated distances.

18. The computer-based system of claim 17, wherein deriving the normalized image based on the image comprises adjusting lighting conditions of the image by one or both of the following:

a histogram equalization method and a local contrast calculation method.

19. The computer-based system of claim 17, wherein representing the rectangular blocks by a DCT comprises, for each rectangular block, steps causing the processors to:

calculate a table of DCT coefficients;

extract the most informative of the DCT coefficients via a zigzag scan of the DCT coefficients of the block starting from a coefficient located at the top-left corner of the table of the DCT coefficients;

replace one or more of the DCT coefficients by weighted differences of corresponding coefficients located in neighboring horizontal and vertical blocks; and

obtain resultant feature values for the block.

20. The computer-based system of claim 17, wherein calculating the distances between the normalized image and the template comprises steps causing the processors to determine a sum of absolute values of the distances between corresponding features of the normalized image and the existing template divided by the sum of the absolute values of the corresponding features.

21. The computer-based system of claim 17, wherein corresponding blocks have the same coordinates.

22. The computer-based system of claim 17, wherein recognizing the normalized image based on the second distance calculated results in a rejection of one or more other templates when a first distance calculated between the normalized image and another template is above a second predetermined threshold.

23. The computer-based system of claim 17, wherein the normalized image is associated with an unknown object and the template image is associated with a first known object and recognizing the normalized image comprises determining a degree of similarity between the normalized image and the template and degrees of similarity between the normalized image and templates associated with any other known object, a degree of similarity being based on a distance between the normalized image and a template.

24. The computer-based system of claim 23, wherein the objects are human faces.