FACE RECOGNITION METHOD

• US 20190311186A1
• Filed: 01/15/2019
• Published: 10/10/2019
• Est. Priority Date: 04/09/2018
• Status: Active Grant

• Current Assignee: Pegatron Corporation
• Original Assignee: Pegatron Corporation
• Inventors: Chen, Po-Sen
• Granted Patent: US 10,922,531 B2
• Time in Patent Office: Days
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 5/046 Forward inferencing; Production systems
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation
• CPC Class Codes: G06V 40/172 Classification, e.g. identification
• CPC Class Codes: G06V 40/173 face re-identification, e.g. recognising unknown faces across different face tracks

Abstract

A face recognition method is disclosed. First, an input image is received. After the input image is received, face recognition is performed on the input image by using a first CNN model to generate at least one first ROI, where each first ROI includes a suspicious image, and a proportion value of a pixel value of the suspicious image in a pixel value of the first ROI is greater than a proportion value of the pixel value of the suspicious image in a pixel value of the input image. Then, face recognition is performed on each first ROI by using a second CNN model to generate at least one second ROI, where the quantity of convolution operation layers of the second CNN model is less than the quantity of convolution operation layers of the first CNN model. Finally, a mark is displayed in the input image.

Claims

1. A face recognition method, comprising:

receiving an input image;

performing face recognition on the input image by using a first convolutional neural network (CNN) model to generate at least one first region of interest (ROI), wherein each first ROI comprises a suspicious image, and a proportion value of a pixel value of the suspicious image in a pixel value of the first ROI is greater than a proportion value of the pixel value of the suspicious image in a pixel value of the input image;

performing face recognition on each first ROI by using a second CNN model to generate at least one second ROI, wherein the quantity of convolution operation layers of the second CNN model is less than the quantity of convolution operation layers of the first CNN model; and

displaying, in the input image, a mark corresponding to the position of the second ROI.

2. The face recognition method according to claim 1, wherein the pixel value of the suspicious image accounts for at least 0.1% of the pixel value of the input image.

3. The face recognition method according to claim 1, wherein the pixel value of the suspicious image accounts for at least 50% of the pixel value of the first ROI.

4. The face recognition method according to claim 1, wherein the first CNN model is established after training is performed by using a plurality of first image materials, the second CNN model is established after training is performed by using a plurality of second image materials, and the plurality of first image materials are different from the plurality of second image materials.

5. The face recognition method according to claim 4, wherein each of the plurality of first image materials comprises a first face image, each of the plurality of second image materials comprises a second face image, the plurality of first image materials has a first pixel value, the plurality of second image materials has a second pixel value, the first face image has a first to-be-detected pixel value, the second face image has a second to-be-detected pixel value, and a first to-be-detected image proportion value of the first to-be-detected pixel value in the first pixel value is different from a second to-be-detected image proportion value of the second to-be-detected pixel value in the second pixel value.

6. The face recognition method according to claim 5, wherein the first to-be-detected image proportion value is less than the second to-be-detected image proportion value.

7. The face recognition method according to claim 6, wherein a proportion value of the first to-be-detected pixel value in the first pixel value is greater than or equal to 0.1%, and a proportion value of the second to-be-detected pixel value in the second pixel value is greater than or equal to 50%.

8. The face recognition method according to claim 7, wherein the plurality of first image materials and the plurality of second image materials are images, of a same face or difference faces, captured in different shadows, at different angles, in different backgrounds, in different light brightness degrees, in different expression changes, or in different shielding degrees.

9. The face recognition method according to claim 1, wherein the input image is a frame of a picture or a film.

10. The face recognition method according to claim 1, wherein the first CNN model comprises a plurality of first convolution operation layers, the second CNN model comprises a plurality of second convolution operation layers, an output of each first convolution operation layer is an input of another subsequent first convolution operation layer, and an output of each second convolution operation layer is an input of another subsequent second convolution operation layer.

11. The face recognition method according to claim 10, wherein an operation parameter quantity of the second convolution operation layer of the second CNN model is 20% of an operation parameter quantity of the first convolution operation layer.

12. The face recognition method according to claim 10, wherein the first convolution operation layer and the second convolution operation layer each comprise a feature extraction layer, and inputs of neurons at the feature extraction layer is connected to a local receptive field at a previous layer to extract a related local feature.

13. The face recognition method according to claim 10, wherein the first convolution operation layer and the second convolution operation layer each comprise a feature mapping layer, the feature mapping layer is obtained by mapping a plurality of features, each mapping is a plane, and weights of neurons at each plane are the same.

14. The face recognition method according to claim 1, wherein the first CNN model further comprises a first pooling operation layer for a pooling operation, and the second CNN model further comprises a second pooling operation layer for a pooling operation.