A Construction Method of Optimal Coding Dictionary for Collaborative Representation
Mei Song-qin · Computer Technology and Development · 2014
In the small size of training samples,the recognition rate of Collaborative Representation( CR) often cannot reach the established expectations. To this end,a construction method of optimal coding dictionary for CR is proposed to improve the classification ability of CR. First,exploit original training samples per subject to produce pseudo training samples. Then,these original training samples and produced pseudo training samples are mixed together to form the original coding dictionary. Finally,the optimal coding dictionary is selected by using the similarity between the test sample and atoms in the original coding dictionary. Experimental results on three widely used face databases show that the proposed method obtains better recognition rate,about 4% ~ 18% higher than the CR method. Moreover,compared with some other similar methods,the proposed method also gets the good recognition rate.