Research On Face Recognition Based On Low-Rank Representation

The Face Recognition is one of the important research topics in the biometric identification field. It contains profound academic background, covers a very wide range and has a rich market application. It was concerned by the top research institutions and scholars at home and abroad of all ages. The face images can be affected by the external environment, such as obstacle, expression, multi-view and other factors, so there are still many challenges and key technologies to be resolved for the Face Recognition.In the past few decades, a large number of algorithms on the Face Recognition Technology have been put forward, and also accompanying by the deep-study and promoting of the excellent algorithm. The raise and widespread use of Sparse Representation based Classification (SRC) have a major milestone for Face Recognition. But this algorithm has its shortage. The Sparse Representation uses a unit matrix as error dictionary. It not only takes a high computation cost, but also cannot conform to the description of the noise and errors. What’s more, the prerequisite of the Sparse Representation is that the training images need over-complete in the same pose and viewpoint. Unfortunately, with the Non-laboratory realistic condition the face images will be rotated, deformed, and corrupted, when they are shot in the different viewpoint and lights. It makes the training images not over-complete and even scarce. Therefore, we mainly focus on the research of Low-rank Representation theory. This paper aims to obtain effective feasible algorithms to improve the the efficiency and accuracy of face recogition. The main contributions of this paper including:(1) This paper comprehensively analyzes the research background, significance, status and the opportunities and challenges of the Face Recognition.(2) By introducing the theories of Sparse Representation based Classification and Collaborative Representation based Classification (CRC), we elucidate the algorithms of the two theories.(3) Inspired by the theory of Low-rank Matrix Recovery, we propose two new Face Recognition algorithms called Robust Principal Component Analysis with Spare Representation based Classification (RPCA_SRC) and Robust Principal Component Analysis with Collaborative Representation based Classification (RPCA_CRC). RPCA_SRC improves the recognition accuracy. And Compared with SRC, RPCA_CRC is much more efficient and effective with less training images.(4) By investigating the Low-rank matrix showed in the texture images. We builds a mathematical model to describe the invariant texture by the rank of the matrix. In the past two decades, numerous "invariant" features and descriptors have been proposed and studied. But they almost focus on extracting various types of feature points or salient regions in the images. So their applications are limited. We propose a robust definition to describe the invariant texture. Based on the theory of Low-rank Representation, we extract certain invariant structures in 3D through their 2D images. In other words, by inverse process the spatial transformation, the rank of texture matrix will be lower, which will make the training matrix have the low-rank structure.