Research On Face Anti-spoofing And Face Recognition System Based On Deep Learning

In recent years,face recognition technology has been widely used,bringing convenience to people's lives.The photo or video of the face is easy to be obtained by others,to ensure the validity of the recognition result,it is necessary to perform a face anti-spoofing to determine whether the face in front of the camera is from a live face.Many face recognition application scenarios are in low-light environments,and traditional face anti-spoofing algorithms are poorly adaptable to low-light environments.Therefore,this paper establishes a face anti-spoofing algorithm in low-light scenes based on deep learning to enhance low-light images,the extraction of facial features in the image and the processing algorithms based on facial features to achieve effective face anti-spoofing for low-light facial images.The face anti-spoofing algorithm in this paper is mainly composed of three parts: low-light image enhancement module,face image feature extraction module,and feature processing module.In the low-light image enhancement module,an image enhancement algorithm based on Retinex theory and deep learning is designed,because it is difficult to extract facial features directly from the image extracted by the camera in the low-light scene.This algorithm realizes the functions of noise reduction,reduction of color distortion caused by low light and restoration of image texture,which have laid the foundation for the subsequent extraction of facial features.The facial image feature extraction module mainly realizes the extraction of stable features that can be used for face anti-spoofing in the enhanced face image.Based on the optimized stacked hourglass network,the module extracts features such as the coordinates of key points such as the corners of the eyes and nose of the face.The features have high reliability and are beneficial to the next stage of processing.The detection of key points of the face by this module is also the basis of face recognition.In the feature processing module,the algorithm mainly processes the key point positions of the face images collected from two cameras from different angles to draw a conclusion about whether it is a living face.The main judgment basis is that the live face has a three-dimensional structure,but the face image in the photo and video playback attack method is a two-dimensional image.Based on this,it is judged whether there is a three-dimensional live face in front of the camera.The actual low-light scene test shows that the accuracy of the low-light face anti-spoofing algorithm proposed in this paper can reach 93.1%.