The Fusion of SRC and SRRC Algorithms

Abstract

As a recently proposed technique, sparse representation based classi-fication(SRC) and sparse residue representation classification SRRChave been widely used for face recognition(FR).SRC and SRRC represent the test sample as a linear combination of training samples. SRC first calculates the coefficient solution via -minimization, and then we calculate the reconstruction residue errors of the test sample generated from each class respectively. However, the SRRC algorithm first exploits the training samples to constructs the error-free samples of the test sample and then calculates the residue between the error-free sample and original test samples. The residue coefficients are also solved by -minimization. In order to integrate the advantages of SRC and SRRC, we use the score level fusion to combine the residues of SRC and SRRC and the test sample is classified into the class. The minimum final residual is the experiment result of the test sample. Compared with previous methods, the fusion algorithm has very competitive FR accuracy and well performance in robustness to occlusion.KeywordsSparse RepresentationSRCSRRC