A Collaborative Sparse Representation Based Approach For Pattern Recognition

The sparse representation classification algorithm has made great breakthroughs in the field of pattern recognition and has been successfully applied to face recognition.The face recognition algorithm based on sparse representation is a concise and efficient method.It is robust to the real situation such as occlusion,lighting and noise.The algorithm has achieved good result in experiments.The face recognition algorithm based on probabilistic collaborative representation is improved on the basis of the sparse representation classification method,which uses the weighted least square algorithm to expand the differences between different classes,but this method still has some disadvantages.On the one hand,the representation of the objective function is not sparse enough,and the convergence speed of this method is slow to some extent.On the other hand,the use of training samples as a dictionary leads to the inability of this method to effectively represent the test samples and is not conducive to using the information hidden between the training samples.In view of the above problems,this paper improves the algorithm from two aspects:improve the accuracy of face recognition by expanding the differences between classes and reducing the differences within classes.The first aspect is improved from the sparse representation.Based on the probabilistic collaborative representation algorithm,the proposed method is more sparse by improving the objective function.On the basis of increasing the weight,the l1 norm term is added.Combining the advantages of probabilistic collaborative representation and sparse representation effectively improves the accuracy of face recognition.The second aspect is improved from dictionary learning.Based on dictionary pair learning and probabilistic collaborative sparse representation algorithms,a collaborative sparse representation classification algorithm based on dictionary learning is proposed.Experiment shows that the improved method proposed in this paper has good recognition results in AR and YALE face databases,and is robust under occlusion and pixel destruction conditions.