Research On Face Recognition In The Presence Of Illumination Variance And Occlusion

Face recognition is an active topic for researchers in the field of pattern recognition, it has broad application prospects. As a biometric identification technology, face recognition has many advantages, such as convenient, non-contact and non-invasive. Face recognition has achieved satisfactory results under controlled environment. However, face images are always under variation of illumination and occlusion in the real-world applications, which brings a serious challenge to face recognition techniques.Based on sparse representation and robust principal component analysis, our research focuses on face recognition with illumination variation and occlusion, the main research results are summarized as follows:1. In this paper, we first introduce the research background and significance of face recognition both at home and abroad. Then we analyze and compare the algorithms dealing with illumination changes and occlusion. After that, we introduce the sparse representation and low-rank matrix recovery in detail, and give some classical methods to solve them.2. Based on low-rank and p-norm sparse prosperities, a matrix regression algorithm for face recognition is proposed. To ensure low rank and sparse prosperities of the matrix regression model, we use low rankness to constraint the regression error, and use thep-norm to constraint the regression coefficients in order to guarantee the the coefficients are close to the sparest solution. Our model uses the alternating direction method to calculate the regression coefficients. Experiment results on face database of AR and Extended Yale B show that the face recognition method proposed in this paper has a higher recognition rate then the current regression methods. And our method is more powerful for removing the structural noise caused by occlusion,and more robust for alleviating the effect of illumination.3. An algorithm which combined robust principal component analysis and low-rank projection for face recognition is proposed. We divide each type of training sample into sum of low-rank matrix and sparse error matrix by RPCA, and construct low-rank projection between training face matrix and the separated low-rank matrix. Then any test face image can obtain a low-rank matrix and a sparse error matrix corresponding to each face category by low-rank projection. In order to obtain discriminating information of the sparse error matrix, we calculate smoothness and do edge detection for the sparse error image, what’s more, we make sum of weighted smoothness and edge information a criterion for classification. Experiment results on face database of AR and Extended Yale B testify the effectiveness of the proposed method with an improved recognition rate.