FACE RECOGNITION USING GRADIENT BASED FEATURE ANALYSIS

• US 20160132718A1
• Filed: 11/06/2014
• Published: 05/12/2016
• Est. Priority Date: 11/06/2014
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Je, Hongmo, Park, Minje, Lee, Hyungsoo, Kim, Tae-Hoon, Cheon, Yeongjae, Kim, Taesup
• Granted Patent: US 9,384,385 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06V 10/243 by compensating for image skew or non-uniform image deformations
• CPC Class Codes: G06V 40/165 using facial parts and geometric relationships

Abstract

Computer-readable storage media, computing devices and methods are discussed herein. In embodiments, a computing device may be configured to perform facial recognition based on gradient based feature extractions of images of faces. In embodiments, the computing device may be configured to determine directional matching patterns of the images from the gradient based feature extraction and may utilize these directional matching patterns in performing a facial recognition analysis of the images of faces. Other embodiments may be described and/or claimed.

Claims

1. An apparatus for computing, comprising:

one or more computing processors;

physical memory coupled with the one or more processors;

a facial recognition module to be loaded into the physical memory and executed by the one or more processors, to;

receive a reference facial feature vector associated with a reference image having a face corresponding to an identifier of a person, wherein the reference facial feature vector includes directional matching patterns of each portions of a grid of a rectangle that encloses a face in the reference image;

receive a test facial feature vector associated with an operational image, wherein the test facial feature vector includes directional matching patterns of each portions of a grid of a rectangle that encloses a face in the operational image;

analyze the reference facial feature vector against the test facial feature vector of the operational image to determine a level of similarity between the directional matching patterns represented by the reference facial feature vector and the directional matching patterns represented by the test facial feature vector, the analysis further includes to;

compare each portion of the operational image grid with a corresponding portion of the reference facial feature vector;

identify one or more portions of the operational image grid that have directional matching patterns similar to the corresponding portion of the reference facial feature vector;

generate a directional histogram for each identified portion of the one or more portions of the grid;

concatenate each directional histogram generated into a histogram of local matching directions for the image; and

identify the face of the image as the person associated with the identifier, in response to determination that the level of similarity between the directional matching patterns represented by the reference facial feature vector and the directional matching patterns represented by the test facial feature vector is above a threshold of similarity.

2. The apparatus of claim 26, wherein the facial feature extraction module is to perform a gradient based feature extraction of a face via:

identification of a rectangle that encloses the face;

detection of a first position of a left eye of the face within the rectangle and a second position of a right eye of the face within the rectangle; and

normalization of the face to cause the left eye to relocate from the first position to a first reference position and the right eye to relocate from the second position to a second reference position.

3. The apparatus of claim 26, wherein the facial feature extraction module, as part of normalization of the face, rotates the image to align the face vertically or resizes the image to enlarge the face.

4. The apparatus of claim 26, wherein the facial feature extraction module, as part of the performance of the gradient based feature extractions, is further to:

divide the rectangle that encloses the face into a plurality of portions to form a grid over the face; and

extract gradient based features from each portion of the grid to determine directional matching patterns within the grid, wherein the directional matching patterns of each portion of the grid are represented by the facial feature vector of the face.

5. The apparatus of claim 4, wherein the facial feature extraction module, as part of the extraction of gradient based features of the plurality of portions, is to perform a histogram of oriented gradients (HOG) analysis of each portion of the plurality of portions of the image, or perform scale invariant feature transformation (SIFT) analysis of each portion of the plurality of portions of the image.

8. The apparatus of claim 1, wherein the facial recognition module, as part of the facial recognition analysis is to further identify the face of the image as that of the person associated with the identifier of the reference facial feature vector, utilizing a binary classification system applied to the histogram of local matching directions.

9. The apparatus of claim 1, wherein the facial recognition module is further to:

determine that a level of similarity between a test facial feature vector of an operational image of and a reference facial feature vector of a reference image is above a predefined threshold of similarity; and

add the operational image of a person to a gallery of images of the person, based on the determination.

10. The apparatus of claim 9, wherein the predefined threshold of similarity is a first predefined threshold of similarity, and wherein to determine that the level of similarity is above the first predefined threshold of similarity further includes determination that the level of similarity is below a second predefined threshold similarity.

11. The apparatus of claim 9, wherein to add the operational image to the gallery of images is further based on one or more of:

a distance between eyes of the face in the image;

alignment of the face in the image;

or

frontality of the face in the image.

26. The apparatus of claim 1, the apparatus further comprising:

a facial feature extraction module, to be loaded into the physical memory and executed by the one or more processors, coupled to the facial recognition module, to;

receive operational and reference images having faces, the reference images having corresponding identifiers of persons;

perform gradient based feature extraction of the faces to determine directional matching patterns of the faces; and

generate facial feature vectors that represent the directional matching patterns of the faces; and

a facial registration module, to be loaded into the physical memory and executed by the one or more processors, coupled with the facial feature extraction module to;

store the facial feature vectors of the reference images as reference facial feature vectors associated with the corresponding identifiers of the persons.

6-7. -7. (canceled)

12-13. -13. (canceled)

16-17. -17. (canceled)

19. (canceled)

23-24. -24. (canceled)

27. A computer-implemented method comprising:

receiving, by a computing device, a reference facial feature vector associated with a reference image having a face corresponding to an identifier of a person, wherein the reference facial feature vector includes directional matching patterns of each portions of a grid of a rectangle that encloses a face in the reference image;

receiving, by the computing device, a test facial feature vector associated with an operational image, wherein the test facial feature vector includes directional matching patterns of each portions of a grid of a rectangle that encloses a face in the operational image;

analyzing, by the computing device, the reference facial feature vector against the test facial feature vector of the operational image to determine a level of similarity between the directional matching patterns represented by the reference facial feature vector and the directional matching patterns represented by the test facial feature vector, the analyzing further includes;

comparing, by the computing device, each portion of the operational image grid with a corresponding portion of the reference facial feature vector;

identifying, by the computing device, one or more portions of the operational image grid that have directional matching patterns similar to the corresponding portion of the reference facial feature vector;

generating, by the computing device, a directional histogram for each identified portion of the one or more portions of the grid;

concatenating, by the computing device, each directional histogram generated into a histogram of local matching directions for the image; and

identifying, by the computing device, the face of the image as the person associated with the identifier, in response to determination that the level of similarity between the directional matching patterns represented by the reference facial feature vector and the directional matching patterns represented by the test facial feature vector is above a threshold of similarity.

14. The computer-implemented method of claim 27, wherein performing the gradient based feature extraction of a face further comprises:

identifying, by the computing device, a rectangle that encloses the face;

detecting, by the computing device, a first position of a left eye of the face within the rectangle and a second position of a right eye of the face within the rectangle;

normalizing, by the computing device, the face to cause the left eye to relocate from the first position to a first reference position and the right eye to relocate from the second position to a second reference position;

dividing the rectangle that encloses the face into a plurality of portions to form a grid over the face; and

extracting gradient based features from each portion of the grid to determine directional matching patterns within the grid, wherein the directional matching patterns of each portion of the grid are represented by the facial feature vector of the face.

15. The computer-implemented method of claim 14, wherein extracting gradient based features of the plurality of portions, includes performing a histogram of oriented gradients (HOG) analysis of each portion of the plurality of portions of the image, or performing scale invariant feature transformation (SIFT) analysis of each portion of the plurality of portions of the image.

18. The computer-implemented method of claim 27, wherein performing facial recognition analysis further comprises identifying the face of the image as that of the person associated with the identifier of the reference facial feature vector, utilizing a binary classification system applied to the histogram of local matching directions.

28. One or more non-transitory computer-readable media having a plurality of instructions stored thereon, the instructions, in response to execution by a processor, to cause the processor to:

receive a reference facial feature vector associated with a reference image having a face corresponding to an identifier of a person, wherein the reference facial feature vector includes directional matching patterns of each portions of a grid of a rectangle that encloses a face in the reference image;

receive a test facial feature vector associated with an operational image, wherein the test facial feature vector includes directional matching patterns of each portions of a grid of a rectangle that encloses a face in the operational image;

analyze the reference facial feature vector against the test facial feature vector of the operational image to determine a level of similarity between the directional matching patterns represented by the reference facial feature vector and the directional matching patterns represented by the test facial feature vector, the analyzing further includes;

compare each portion of the operational image grid with a corresponding portion of the reference facial feature vector;

identify one or more portions of the operational image grid that have directional matching patterns similar to the corresponding portion of the reference facial feature vector;

generate a directional histogram for each identified portion of the one or more portions of the grid;

concatenate each directional histogram generated into a histogram of local matching directions for the image; and

identify the face of the image as the person associated with the identifier, in response to determination that the level of similarity between the directional matching patterns represented by the reference facial feature vector and the directional matching patterns represented by the test facial feature vector is above a threshold of similarity.

20. The one or more non-transitory computer-readable media of claim 28, wherein the instructions, in response to execution by a processor, further cause the processor to:

receive reference images having faces and corresponding identifiers of persons;

perform gradient based feature extraction of the faces to determine directional matching patterns of the faces;

generate facial feature vectors that represent the directional matching patterns of the faces; and

store the facial feature vectors of the reference images as reference facial feature vectors associated with the corresponding identifiers of the persons.

21. The one or more non-transitory computer-readable media of claim 20, wherein to perform the gradient based feature extraction of a face further comprises:

identification of a rectangle that encloses the face;

detection of a first position of a left eye of the face within the rectangle and a second position of a right eye of the face within the rectangle;

normalization of the face to cause the left eye to relocate from the first position to a first reference position and the right eye to relocate from the second position to a second reference position;

division of the rectangle that encloses the face into a plurality of portions to form a grid over the face; and

extraction of gradient based features from each portion of the grid to determine directional matching patterns within the grid, wherein the directional matching patterns of each portion of the grid are represented by the facial feature vector of the face.

22. The one or more non-transitory computer-readable media of claim 21, wherein extraction of gradient based features of the plurality of portions, includes performance of a histogram of oriented gradients (HOG) analysis of each portion of the plurality of portions of the image, or performance of a scale invariant feature transformation (SIFT) analysis of each portion of the plurality of portions of the image.

25. The one or more non-transitory computer-readable media of claim 28, wherein to perform facial recognition analysis further comprises:

identification of the face of the image as that of the person associated with the identifier of the reference facial feature vector, utilizing a binary classification system applied to the histogram of local matching directions.