Method, system, and computer program product for recognizing face

Patent metadata

Field	Value
Publication number	US-10262190-B2
Application number	US-201514888360-A
Country	US
Kind code	B2
Filing date	Mar 26, 2015
Priority date	Mar 26, 2015
Publication date	Apr 16, 2019
Grant date	Apr 16, 2019

Abstract

A method, system and computer program product for recognizing a face are provided, comprising: acquiring an image for the face; detecting a set of first feature points representing detail features of the image; extracting, for each first feature point in the set of first feature points, a first descriptor describing feature information on the first feature point; acquiring, for each second feature point in a set of second feature points, a second descriptor describing feature information on the second feature point; detecting matched feature point pairs between the set of first feature points and the set of second feature points, based on the first descriptor and the second descriptor; calculating the number of the matched feature point pairs; and recognizing the image as being consistent with the registered image, if the number of the matched feature point pairs is larger than a first preset threshold.

First claim

What is claimed is:

1. A method for recognizing a face, comprising:
acquiring an image to be recognized, for the face;
detecting a set of first feature points representing detail features of the image to be recognized, in the image to be recognized;
extracting, for each first feature point in the set of first feature points, a first descriptor describing feature information on the first feature point;
acquiring, for each second feature point in a set of second feature points, a second descriptor describing feature information on the second feature point, the set of second feature points representing detail features of a pre-stored registered image;
detecting matched feature point pairs between the set of first feature points and the set of second feature points, based on the first descriptor and the second descriptor;
calculating the number of the matched feature point pairs; and
recognizing the image to be recognized as being consistent with the registered image, if the number of the matched feature point pairs is larger than a first preset threshold,
wherein detecting matched feature point pairs between the set of first feature points and the set of second feature points comprises:
calculating a similarity transformation matrix between the image to be recognized and the registered image based on a set of first key points representing structure information on the image to be recognized and a set of second key points representing structure information on the registered image;
calculating, for each first feature point, a first match region in the registered image with respect to the first feature point based on the similarity transformation matrix;
calculating, for each second feature point in the first match region, a Euclidean distance between the first descriptor of the first feature point and the second descriptor of the second feature point;
determining a relationship between a ratio between a smallest Euclidean distance and a second smallest Euclidean distance, and a third preset threshold; and
determining the first feature point and the second feature point with the smallest Euclidean distance to the first descriptor of the first feature point in the first match region as the matched feature point pair if the ratio between the smallest Euclidean distance and the second smallest Euclidean distance is smaller than the third preset threshold;

or detecting matched feature point pairs between the set of first feature points and the set of second feature points comprises:
calculating a similarity transformation matrix between the image to be recognized and the registered image based on the set of first key points and the set of second key points;
calculating, for each second feature point, a second match region in the image to be recognized with respect to the second feature point based on the similarity transformation matrix; and
calculating, for each first feature point in the second match region, a Euclidean distance between the second descriptor of the second feature point and the first descriptor of the first feature point;
determining a relationship between a ratio between a smallest Euclidean distance and a second smallest Euclidean distance, and a third preset threshold; and
determining the second feature point and the first feature point with the smallest Euclidean distance to the second descriptor of the second feature point in the second match region as the matched feature point pair if the ratio between the smallest Euclidean distance and the second smallest Euclidean distance is smaller than the third preset threshold.

2. The method of claim 1 , wherein a resolution of the image is larger than a preset resolution threshold.

3. The method of claim 1 , further comprising:
detecting a face region image in the image to be recognized;
wherein, detecting the set of first feature points representing detail features of the image to be recognized in the image to be recognized comprises:
detecting the set of first feature points based on the face region image.

4. The method of claim 3 , wherein detecting the set of first feature points comprises:
scaling the face region image into different scales; and
detecting a location and a size of the first feature point using an off-line trained feature point classifier, for the face region image in each scale.

5. The method of claim 3 , wherein detecting the set of first feature points comprises:
performing a convolution process on the face region image with Gaussian cores in different scales, to obtain corresponding Gaussian images in different scales;
performing a differentiating process on the Gaussian images in adjacent scales to obtain Gaussian differentiating images;
determining an extreme point in each Gaussian differentiating image, wherein a value of the extreme point is larger than values of adjacent points in the Gaussian differentiating image and is larger than values of adjacent points in the Gaussian differentiating images in adjacent scales; and
determining an extreme point as the first feature point if the value of the extreme point is larger than a second preset threshold.

6. The method of claim 1 , wherein extracting, for each first feature point in the set of first feature points, a first descriptor describing feature information on the first feature point comprises:
performing a normalization process on each first feature point, with the first feature point as a preset reference, to obtain a feature point region image; and
acquiring the first descriptor of the first feature point using an off-line trained descriptor extractor, for the feature point region image.

7. The method of claim 1 , wherein extracting, for each first feature point in the set of first feature points, the first descriptor describing feature information on the first feature point, comprises:
determining, for each first feature point, a feature point region with the first feature point as a preset reference;
dividing the feature point region into multiple sub regions and calculating gradient information on each sub region;
calculating a multiple dimensional gradient histogram of each sub region based on the gradient information; and
connecting the number of points in each dimension of the multiple dimensional gradient histogram as a feature vector, to obtain the first descriptor of the first feature point.

8. The method of claim 1 , further comprising:
detecting the set of first key points.

9. The method of claim 1 , wherein the second descriptor is obtained by:
acquiring the registered image;
detecting the set of second feature points representing detail features of the registered image, in the registered image;
extracting, for each second feature point in the set of second feature points, a second descriptor describing the feature information on the second feature point; and
storing the second descriptor.

10. The method of claim 1 , wherein calculating the number of the matched feature point pairs comprises:
obtaining a preliminary match result of the matched feature point pairs between the set of first feature points and the set of second feature points, based on the first descriptor and the second descriptor; and
screening the preliminary match result based on a Random Sample Consensus method, to obtain the matched feature point pairs.

11. A system for recognizing a face, comprising:
a processor;
a memory;
computer program instructions stored in the memory, which, when executed by the processor, perform steps of:
acquiring an image to be recognized, for the face;
detecting a set of first feature points representing detail features of the image to be recognized, in the image to be recognized;

Assignees

• Beijing Kuangshi Tech Co Ltd
• Pinhole Beijing Tech Co Ltd

Inventors

• He Tao

Classifications

• G06V40/171 Primary

  Local features and components; Facial parts (eye characteristics G06V40/18); Occluding parts, e.g. glasses; Geometrical relationships · CPC title

• G06V40/168

  Feature extraction; Face representation · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06V10/478

  Contour-based spectral representations or scale-space representations, e.g. by Fourier analysis, wavelet analysis or curvature scale-space [CSS] · CPC title

• G06K9/00288

  Physics · mapped topic