Research On Spoofing Detection And Application In Face Recognition

Face recognition technology has been widely used in military,economic,and public security fields in recent years due to its non-contact and convenient characteristics.At present,most face recognition systems can recognize the identity of the input face image,but cannot accurately distinguish whether the input face belongs to the real face of a legal user or a fake face by an attacker.Nowadays,the Internet is very developed,and attackers can easily obtain real-person face images and videos through various social platforms.Once the attackers use these legitimate users 'face photos and videos to attack the face authentication system,it may cause more serious consequences and losses.Therefore,how to quickly and effectively distinguish the authenticity of the face input by the system to ensure the safety of the face recognition system has become an urgent problem in face recognition technology.In order to solve this problem,face live detection technology came into being,which aims to distinguish whether the face image collected by the face authentication system belongs to a real face image of a legitimate user or a fake face image forged by an attacker.In addition,the face authentication system will be affected by changes in the external environment and the diversity of counterfeiting methods during the process of collecting face images,which also brings serious challenges to the live face detection technology.This paper studies the living body detection technology in the face recognition process.The main research contents and innovations are as follows:1.In the face recognition process,illumination changes will greatly affect the recognition accuracy.To this end,a discrete wavelet transform to enhance contrast limited adaptive histogram equalization(DWT E-CLAHE)illumination preprocessing method is proposed.The algorithm first performs Gamma correction on the original face image;then uses discrete wavelet transform to extract the low-frequency and high-frequency components of the image;finally,the low-frequency components are sequentially applied logarithmic transformation,Gamma correction,and contrast-limited adaptive histogram equalization processing.Thus,the pre-processed face image is obtained.The algorithm has a good pre-processing effect on face images under insufficient lighting and extreme lighting conditions.After testing on face libraries such as AR,CMU PIE and Extend Yale B,the algorithm has high efficiency.2.Aiming at the problems of manual feature design in face detection,single feature extraction and traditional deep learning algorithms that are prone to produce local minimums and over fitting,an extreme learning machine based on local receptive fields(ELM-LRF)face live detection method.The model first randomly generates input weights,and then uses regularized least squaresto analytically calculate the output weights.Generalization performance.On the CASIA-FASD,NUAA,and Print-Attack databases,the ELM-LRF model is compared with other advanced face detection algorithms.The ELM-LRF algorithm has high classification performance.