Face Recognition Based On Sparse Representation Classification

Classification based on Sparse representation for face recognition is an effective face recognition algorithm,It is assumed that a test image can be sparsely represented on the training image,Then calculate the smallest class error to classify,Classification based on Sparse representation is robust to occlusion,illumination and noise,and has a good experimental effect.But it uses the training sample directly as a dictionary,Resulting in a failure to effectively represent the test image.Using the original sample as a dictionary will not be able to take full advantage of hiding information between training samples.Coding method of Classification based on Sparse representation for face recognition generally used is l0 norm approximation norm.But this method also has shortcomings.On the one hand,in some ways,the convergence of these methods is slow.On the other hand,when the test sample is in a very large random occlusion,The sparse solution is so dense that it is difficult to find the correct class of test samples.Based on these two shortcomings,we have improved the algorithm in two aspects.The first improved algorithm is to improve the dictionary,Based on the existing algorithms,a simplified Fisher discriminant dictionary is proposed,The model is simplified by combining the terms of coordination and the discriminative penalty in the fidelity terms,effectively improve the face recognition rate.The second improved algorithm is to improve the sparse coding,On the basis of the iterative reweighted least squares method,we propose a iterative double-weighted least squares method.Comparisons with other l1 minimization and l0 minimization algorithms,And the experimental results on AR,ORL and YALE face databases are presented,The two algorithms in this paper have better reconstruction effect and recognition effect than the existing algorithms.