Living body detection method and system

Abstract

The invention discloses a living body detection method. The method comprises the steps of acquiring a visible light face image; generating an infrared-imitated face image according to the visible light face image; respectively generating a face pseudo-depth image and a face feature image according to the visible light face image and the infrared-imitated face image; fusing the features of the facepseudo-depth image and the features of the face feature image to obtain a capacity and feature image; and generating a detection result according to the fusion feature image. Compared with the priorart, the application has the following beneficial effects: the application has the following beneficial effects, an end-to-end deep learning neural network method is adopted and a GAN generated infrared face technology is provided, visible light image information characteristics and infrared image information characteristics are provided for monocular living body detection, a binocular living body detection framework is simulated, extraction of network layer fusion characteristics of different depths and generation of a pseudo depth map are achieved, meanwhile, a method for inducing GAN to generate true face characteristics and pseudo face characteristics is provided, and the living body detection algorithm effect is improved.

Description

technical field [0001] The present application relates to the field of network security, and in particular, relates to a living body detection method. Background technique [0002] Face liveness detection technology can be divided into cooperative liveness detection and non-cooperative (silent) liveness detection from the detection process. Cooperative liveness detection has high security performance and is mainly used in scenarios with high security levels (such as payment scenarios). However, due to the need for human-computer interaction, the user experience effect is not very good for scenarios with high timeliness. The non-cooperative type is silent and living, and the user experience is good. It is widely used in scenarios with low security levels (such as access control, check-in, etc.). [0003] In terms of sensor types and combinations, human face liveness detection can be divided into visible light (RGB), infrared monocular liveness and combined binocular liveness...

AI Technical Summary

Problems solved by technology

Based on monocular living detection schemes (such as monocular RGB, monocular infrared), due to their single input source, the anti-interference performance and accuracy of complex scenes such as light changes, distances, and face angle changes are poor; structured light-based Liveness detection has better performance, but it is difficult to cover more practical application scenarios due to factors such as high price and short recognition distance.

Based on the binocular living body detection algorithm, at present, the depth information is mainly estimated through the binocular image for face live body recognition, but the recognition effect for 3D face masks is not good.

Images