基于稀疏表示的人脸识别的应用研究-控制工程专业论文.docx

万方数据 万方数据 Abstract Face recognition is one of the research focuses among the area of image processing. And it has already been widely used among many areas such as pattern recognition, artificial intelligence, and computer vision and so on. Sparse representation based face recognition, with a simple theoretical basis and a robust performance, has attracted close attention from many researchers. After studying many related articles, this thesis has found that it is the face occlusion, illumination variation, and poor face alignment that stop the face recognition from practical appliance. This thesis is aimed to handle these problems. The mainly researching contents in this thesis conclude: This thesis starts from introducing the theory of sparse representation. According to the theory, this thesis describes a traditional scheme which is called face recognition via sparse representation based classification. Then, this thesis provides a new scheme called face recognition via robust sparse representation, which is more robust to outliers in the image. The face recognition via robust sparse representation algorithm has a good performance in dealing with face occlusion and illumination variation. And this thesis introduces the use of random features into the face recognition scheme. In order to handle the poor face alignment problem, this thesis makes use of a matrix rank minimization based face alignment model. For the sake of solving the linearized convex optimization problem in the face alignment model, this thesis proposes an algorithm called the inexact augmented Lagrange multipliers algorithm which has a good performance. This thesis will carry out experiments between proposed algorithms and the popular PCA, LDA algorithms on LFW, ORL and Yale database. Based on the experiment results, this thesis finds that sparse representation based face recognition algorithms are better than the popular PCA algorithm and LDA algorithm. And when there are outliers in the face im