Face silent living body detection method and system based on multi-feature extraction module

Abstract

The invention relates to a face silent living body detection method and system based on a multi-feature extraction module, and the system comprises a texture feature extraction module which is used for extracting the texture features of a living body target and a non-living body target; extracting different levels of texture features of the target; the space attention module is used for refining texture features; the non-living body feature extraction module is used for amplifying the feature difference between a living body sample and a non-living body sample; the feature map refining module is used for generating a feature map; and the living body distinguishing module is connected to the feature map refining module and is used for distinguishing a living body. According to the method, manual pixel-level labeling is not needed in the training stage, the labor cost is saved, feature differences between the living body samples and between the living body samples and the non-living body samples are supervised and compared through metric learning, and clues for pixel-level supervision of the images are automatically mined.

Description

technical field [0001] The invention relates to the technical field of computer target detection, in particular to a face silent living body detection method and system based on a multi-feature extraction module. Background technique [0002] With the development of deep learning technology, face-related computer vision technology continues to mature, and methods based on deep learning have achieved good results in the living field. [0003] Liveness detection algorithms can be divided into cooperative liveness detection and non-cooperative liveness detection according to different requirements: [0004] (1) Cooperative liveness detection requires the user to make corresponding actions according to the instructions, and judge whether the detected user is a real living body by verifying whether the user's actions are consistent with the order of the instructions. [0005] (2) Non-cooperative liveness detection is also called silent liveness detection. The user only needs to ...

AI Technical Summary

Problems solved by technology

[0008] In the prior art, the central difference convolution has been used for feature extraction, and the classification score of the generated feature map is compared with the preset threshold to judge whether the input is a living body; but it has not been used for multi-feature extraction, and it cannot be judged non-living target

[0009] However, the current silent living detection method for RGB images has the disadvantages of low detection accuracy in practical applications, and the need to train additional modules to generate auxiliary supervision information and perform pixel-level supervision on images.

Images