A weighted sparse neighbor representation based on Gaussian kernel function to face recognition

Currently, sparse representation was widely used in face recognition. However, traditional sparse representation method cannot effectively consider the effect of different weight of training samples When reconstruct the test samples. In this paper, a weighted sparse neighbor representation based on Gaussian kernel function model is presented to resolve above problems. Firstly, K nearest training samples is selected for constructing a new training dictionary according to the Euclidean distances between the test samples and training samples. Then, a weight is given to each sparse coefficient of new training sample. Above sparse coefficient is solved by norm L1 minimization method. Finally, recognition task is performed by the minimum reconstruction error of sparse coefficient. Experimental results illustrate that, the proposed algorithm achieves 96.64% correct recognition rate, which is significantly higher than the various existing comparison methods.