Research On Face Recognition Algorithm Based On RPCA And Deep Learning

Face recognition is a major research topic in the field of big data analysis and information processing.Deep learning uses large-scale neural network models to extract facial features,which has good recognition accuracy compared with classical face recognition methods.Robust principal component analysis?RPCA?can effectively pre-process noise-contaminated face data to reduce the impact of noise.This paper mainly studies the two methods in the face recognition process,because the face data is susceptible to illumination,noise,occlusion and other factors affecting the recognition rate.The main research work is as follows:?1?A face recognition algorithm based on RPCA and local binary pattern?LBP?and deep belief network?DBN?is proposed to solve the problem that the illumination shadow caused by illumination changes is poorly recognized in LBP+DBN.Firstly,the convex optimization model of RPCA is solved by the inexact augmented lagrange multiplier method to obtain the low rank image of the sample.Then the feature extraction of the face image and the test image of the low rank structure is performed by LBP.Finally,construct a DBN model for classification and identification.Experiments on Extended YaleB face database show that the proposed algorithm is more robust than LBP method,LBP+DBN method and low rank recovery method.?2?A face recognition algorithm based on RPCA and robust local binary pattern?RLBP?is proposed to reduce the influence of residual noise on feature extraction in low rank recovery with large noise images.Firstly,the RPCA convex optimization model is used to low-rank recovery the face image with noise by augmented lagrange multiplier method.Then the RLBP is used to extract the features of the low rank recovery image.Finally,the deep belief network is used for classification and recognition.Experiments are carried out by adding different degrees of noise to the Extended YaleB face database.Experiments show that the proposed algorithm has a higher recognition rate than the LBP+DBN method and RPCA+LBP+DBN method.It has good robustness and a certain improvement in recognition efficiency.?3?A face recognition algorithm based on L₂₁ norm RPCA and convolutional neural networks?CNN?is proposed,which solves the problem that the traditional RPCA is not ideal for occlusion of low-rank recovery of face images.Firstly,based on the L₂₁ matrix norm,the characteristics of the matrix row and column information can be fully embodied.The L₂₁-RPCA is used to perform low-rank recovery on the occluded face sample data,and then the low-rank image is used as the input of the convolutional neural network for convolution training.Finally,Classify and identify using the Softmax classifier.Experiments are carried out by adding different sizes of occlusion blocks in the Extended YaleB face database.Experiments show that the algorithm has better advantages than the traditional CNN method and RPCA method.