Face Recognition Based On Feature Learning

With the arrival of the information age, auto face recognition has become a hot topic of pattern recognition and computer vision. As a special biometric technology, face recognition technology has been applied to the area of social public security and our daily life. However, there are still many key issues to be solved in face recognition, especially how to extract the most effective features.This paper studies several classic face recognition methods. Inspired by the classic algorithms, we work out some new algorithms which can identify faces efficiently. The following are the main work and contributions of this paper:(1) This paper makes a brief introduction and summary of the methods currently used in face recognition technology. We introduce the main pretreatment methods for lighting transformation in detail and make some experiments on Yale B & Extended Yale B database at the same time.(2) A novel face recognition method named MPCA on Gabor Tensor is proposed. The basic idea of the method is convolving the Gabor wavelet with the original image, and then pictures the forty images as a tensor object. After that, we transform the high-dimensional feature space into low-dimensional subspace using MPCA. The proposed method can better preserve the spatial structure compared with the traditional methods such as PCA and GPCA. Experiment results show that our method is robust to the illumination, posture and expression changes to some degree.(3) This paper presents a structured low-rank representation based on projection method. The basic idea of the low-rank representation is to represent the test sample using the linear combination of all the training samples. The proposed method can transform the original image to a low feature subspace and preserve the low-rank structure property of the samples. In addition, an improved structured low-rank representation based on projection method was proposed with the Gabor tensor features. Experimental results on multiple face database show that the method is robust to illumination, posture and expression changes to some degree.