Research On Face Presentation Attack Detection Based On Illumination Consistency And Texture Feature

With the development of the new generation of information technology,face recognition has been widely used in various authentication scenarios.At the same time,the security issue of face recognition has also caused strong concern and has become the main attack target.Therefore,in order to improve the security of face recognition systems,face presentation attack detection emerges and becomes an indispensable part.This paper analyzes the problems existing in the existing research on face presentation attack detection,introduces the common face spoofing attack methods,and analyzes the difference between live faces and prosthetic faces.Taking illumination consistency and texture features as the starting point,the face presentation attack detection algorithms based on deep learning are studied.The main innovations in this paper are summarized as follows.(1)Aiming at the performance reduction of the face presentation attack detection under complex illumination conditions,a face presentation attack detection algorithm based on two-stream Vision Transformers fusion is proposed.The algorithm designs a two-stream Vision Transformers model framework for migration learning,which is divided into two channels(RGB stream and MSRCR stream)to extract complementary depth features.In order to obtain fusion features that take into account texture discrimination and illumination robustness,a feature fusion method based on Self-attention is proposed.The experimental results show that it has good robustness and generalization performance when facing different lighting and background conditions.(2)In order to make full use of the spatial details of the image while taking into account the cost,a face presentation attack detection algorithm based on texture gradient enhancement and multiscale fusion is presented.In order to obtain features with more texture discrimination,a texture gradient enhancement method is designed.Subsequently,features containing rich contextual information are obtained at low cost through the designed multi-scale fusion method.The experimental results show that the algorithm achieves good singledatabase performance in intra-dataset experiments and good generalization performance in cross-dataset experiments.(3)A face presentation attack detection system is designed and implemented,which provides a good experimental platform for related research work.The algorithms proposed in this paper is easy to deploy,fast to detect,and does not require user-specific cooperation and direct contact with devices.It shows good application prospects in ensuring the security of a face recognition system.