The Research Of Sparse Representation Face Recognition Algorithms In Uncontrolled Environment

Face recognition has wide application value and research significance, which is the research hotspot in the field of biological recognition and computer vision. This paper studied the sparse representation algorithms in uncontrolled environment. The research work is as follows:(1) The algorithm of the sparse representation based on HOG features and polynomial kernel function was proposed. The HOG features are robust to geometric and optical deformation. The features will stay robust when illumination and expression change. But its dimension is very high. Nuclear method gives a good way to solve the problem. The features from low-dimensional input space are mapped to high-dimensional feature space, making them linearly separable in the space.(2) A uniform local binary pattern and sparse representation face recognition algorithm based on Gabor phase and amplitude was proposed. Gabor wavelet kernel function has the same characteristics as simple cells of human cerebral cortex in 2D reflection area, which is able to caputure images’spatial frequency, spatial location and direction information. Meanwhile, uniform loacal binary pattern is robust to the variatious of illumination. In this proposed algorithm, the Gabor phase and amplitude images of a face image are obtained by using Gabor filter. Then uniform local binary histogram is extracted via block. Finally, the test image can be classified as the existing class via sparse representation.(3) Face Recognition algorithm based on discriminative dictionary learning and regularized robust coding was proposed. The fisher discrimination criterion is imposed on the coding coefficients so that they have small within-class scatter and big between-class scatter. In this proposed algorithm, the Gabor amplitude images of a face image are obtained by using Gabor filter. Then uniform local binary histogram is extracted. A new dictionary is gained by using the fisher criterion. Finally, the test image can be classified as the existing class via regularized robust coding.(4) Regularized robust coding for face recognition based on a new dictionary was proposed. Firstly, a binary image is gained by gray threshold transformation and a more clear image without some isolated points can be obtained with smoothing effect. Sencondly, a new dictionary can be obtained in the way of fusing the binary image and the original training dictionary. Finally, the test image can be classified as the existing class via regularized robust coding. The experimental results based on AR face database show that the proposed algorithm has higher face recognition rate compared with regularized robust coding for face recognition.