Face authentication using recognition-by-parts, boosting, and transduction

• US 8,194,938 B2
• Filed: 06/01/2010
• Issued: 06/05/2012
• Est. Priority Date: 06/02/2009
• Status: Active Grant

• Current Assignee: George Mason Intellectual Properties, Inc.
• Original Assignee: George Mason Intellectual Properties, Inc.
• Inventors: Wechsler, Harry, Li, Fayin
• Primary Examiner(s): Bali, Vikkram
• Application Number:
  US12/791,162
• Publication Number:
  US 20110135166A1
• Time in Patent Office: 735 Days
• Field of Search: None
• US Class Current: 382/118
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/2413 based on distances to training or reference patterns
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/774 Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 40/171 Local features and components; Facial parts eye characteristics G06V40/18; Occluding parts, e.g. glasses; Geometrical relationships
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

A robust recognition-by-parts authentication system for comparing and authenticating a test image with at least one training image is disclosed. This invention applies the concepts of recognition-by-parts, boosting, and transduction.

Claims

1. A robust recognition-by-parts face authentication system for determining if at least one query image obtained using an imaging device matches at least one training image in an enrollment gallery comprising:

a. an enrollment module comprising;

1) an enrollment patch extractor configured for extracting a multitude of training patches at different scales for each center position of the training image;

2) an enrollment patch processor configured for;

a) selecting the training patches that are predictive in identifying the training image; and

b) reducing the selected training patches'"'"' dimensionality using transduction;

3) an enrollment part clustering module configured for clustering the selected training patches into training exemplar-based parts for matching and authentication using K-means;

4) an enrollment data fusion module configured for enrolling the training exemplar-based parts using boosting and transduction;

b. a query module comprising;

1) a query patch extractor configured for extracting a multitude of query patches at different scales for each center position of the query image;

2) a query patch processor configured for;

a) selecting the query patches that are predictive in identifying the query target; and

b) reducing the selected query patches'"'"' dimensionality using transduction;

3) a query part clustering module configured for clustering the selected query patches into query exemplar-based parts for matching and authentication using K-means; and

c. an ID authentication module configured for matching the query exemplar-based parts against a gallery of all the enrolled training exemplar-based parts using flexible matching.

2. The robust recognition-by-parts face authentication system according to claim 1, wherein the enrollment patch extractor uses SIFT for extracting the training patches.

3. The robust recognition-by-parts face authentication system according to claim 1, wherein the enrollment patch extractor uses Gabor wavelet for extracting the training patches.

4. The robust recognition-by-parts face authentication system according to claim 1, wherein the query patch extractor uses SIFT for extracting the query patches.

5. The robust recognition-by-parts face authentication system according to claim 1, wherein the query patch extractor uses Gabor wavelet for extracting the query patches.

6. The robust recognition-by-parts face authentication system according to claim 1, further including a training image preprocessor module configured for preprocessing at least one training image by:

a. converting 256 gray levels into floating points;

b. using geometric normalization that lines up chosen eye coordinates;

c. cropping the training image using an elliptical mask;

d. equalizing a histogram of the training image; and

e. normalizing pixel values to mean zero and variance of one.

7. A robust recognition-by-parts face authentication device for determining if at least one query image obtained using an imaging device matches at least one training image in an enrollment gallery comprising:

a. an enrollment module comprising;

1) an enrollment patch extractor configured for extracting a multitude of training patches at different scales for each center position of the training image;

2) an enrollment patch processor configured for;

a) selecting the training patches that are predictive in identifying the training image; and

b) reducing the selected training patches'"'"' dimensionality using transduction;

3) an enrollment part clustering module configured for clustering the selected training patches into training exemplar-based parts for matching and authentication using K-means;

4) an enrollment data fusion module configured for enrolling the training exemplar-based parts using boosting and transduction;

b. a query module comprising;

1) a query patch extractor configured for extracting a multitude of query patches at different scales for each center position of the query image;

2) a query patch processor configured for;

a) selecting the query patches that are predictive in identifying the query target; and

b) reducing the selected query patches'"'"' dimensionality using transduction;

3) a query part clustering module configured for clustering the selected query patches into query exemplar-based parts for matching and authentication using K-means; and

c. an ID authentication module configured for matching the query exemplar-based parts against a gallery of all the enrolled training exemplar-based parts using flexible matching.

8. The robust recognition-by-parts face authentication device according to claim 7, wherein the enrollment patch extractor uses SIFT for extracting the training patches.

9. The robust recognition-by-parts face authentication device according to claim 7, wherein the enrollment patch extractor uses Gabor wavelet for extracting the training patches.

10. The robust recognition-by-parts face authentication device according to claim 7, wherein the query patch extractor uses SIFT for extracting the query patches.

11. The robust recognition-by-parts face authentication device according to claim 7, wherein the query patch extractor uses Gabor wavelet for extracting the query patches.

12. The robust recognition-by-parts face authentication device according to claim 7, further including a training image preprocessor module configured for preprocessing at least one training image by:

a. converting 256 gray levels into floating points;

b. using geometric normalization that lines up chosen eye coordinates;

c. cropping the training image using an elliptical mask;

d. equalizing a histogram of the training image; and

e. normalizing pixel values to mean zero and variance of one.

13. A physical and tangible computer readable medium encoded with instructions for determining if at least one query image obtained using an imaging device matches at least one training image in an enrollment gallery, wherein execution of the instructions by one or more processors causes the one or more processors to perform the steps comprising:

a. extracting a multitude of training patches at different scales for each center position of the training image;

b. processing the training patches by;

1) selecting the training patches that are predictive in identifying the training image; and

2) reducing the selected training patches'"'"' dimensionality using transduction;

c. clustering the selected training patches into training exemplar-based parts for matching and authentication using K-means;

d. enrolling the training exemplar-based parts using boosting and transduction;

e. extracting a multitude of query patches at different scales for each center position of the query image;

f. processing the query patches by;

1) selecting the query patches that are predictive in identifying the query target; and

2) reducing the selected query patches'"'"' dimensionality using transduction;

g. clustering the selected query patches into query exemplar-based parts for matching and authentication using K-means; and

h. matching the query exemplar-based parts against a gallery of all the enrolled training exemplar-based parts using flexible matching.

14. The physical and tangible computer readable medium according to claim 13, wherein SIFT is used for extracting the training patches.

15. The physical and tangible computer readable medium according to claim 13, wherein Gabor wavelet is used for extracting the training patches.

16. The physical and tangible computer readable medium according to claim 13, wherein SIFT is used for extracting the query patches.

17. The physical and tangible computer readable medium according to claim 13, wherein Gabor wavelet is used for extracting the query patches.

18. The physical and tangible computer readable medium according to claim 13, further including preprocessing at least one training image by:

a. converting 256 gray levels into floating points;

b. using geometric normalization that lines up chosen eye coordinates;

c. cropping the training image using an elliptical mask;

d. equalizing a histogram of the training image; and

e. normalizing pixel values to mean zero and variance of one.