Modularized dynamic configurable living body face recognition system

Abstract

The invention relates to a modularized dynamic configurable living body face recognition system, which comprises an image input module, a transmission module, a living body detection module realized by adopting a fusion feature convolutional network, a face detection module realized by adopting a face feature extraction convolutional neural network, a face recognition module and a result output module. Under different illumination intensities, a higher face recognition result is obtained, and under the conditions of no light, weak light and normal light sources, the feature fusion network has high robustness and high accuracy.

Description

technical field [0001] The present invention relates to a technology in the field of neural network application, specifically a modular dynamic configurable living body face recognition system, which utilizes multimodal data to improve the accuracy of silent living body detection, and is suitable for color, near-infrared , sound information, depth information, action timing information and other multi-modal information. Background technique [0002] The existing 2D face recognition system is easy to be attacked by criminals. The common attack types include printing plane attack, screen replay attack, 3D mask attack, etc. If you don’t judge the malicious attack here first, you can directly conduct human Face recognition will greatly reduce the reliability of the face recognition system. The existing face recognition methods use the Retinex algorithm to calculate the degree to which the face image is affected by visible light, adjust the light and dark values ​​of the face im...

AI Technical Summary

Problems solved by technology

This technology obtains color images and infrared depth images, detects faces through deep learning models on color images, analyzes the screen pixel texture of the face area to be tested, and uses the Surf algorithm to determine whether the face area to be tested meets the preset conditions , carry out live face judgment on the face area and anti-attack judgment of face live body and face three-dimensional mask, but this existing technology cannot correct in time when the face moves, which will lead to the difference between the front and back pictures of the face area Inconsistency, which in turn affects the subsequent liveness detection function

Using the pixel points with depth information in the face area to judge whether the proportion of the face area of ​​the infrared depth image exceeds the preset threshold will result in: the face area may have a hollow area due to the occlusion problem during the imaging process of the depth camera, Real people cannot pass; the preset threshold depends on a specific data set, which cannot achieve good model generalization and the problem that 3D face masks and head models will be missed

In addition, the usability of living body detection judgment based on whether there is a frame in the face area is extremely low. The area of ​​the remake attack may cover the entire imaging area of ​​the camera and the frame cannot be captured in the image. intercepted by mistake

Images