A hierarchical discriminative canonical correlation analysis based on the local sparse class representation

Category information and hidden structure information of the data, For the recognition effect affects significantly, The CCA does not consider the category and structural information of the sample, therefore, Identification effect is often poor, To improve the recognition effect, Better use of category information and structural information, Many variants of CCA have successively, DCCA adds category information in the framework of CCA, However, ignoring structural information, The LDCCA considers both category information and structural information, but, The determination of the local nearest neighbor k is a puzzle, To address the above deficiencies, This paper presents canonical correlation analysis algorithms based on local sparse class representation and hierarchical discrimination, By the method of hierarchical identification, Comprehensive use of the local and overall discriminant information, Shorboard that improve local discrimination methods are sensitive to the size of neighborhood k. Improve the identification ability of the extraction features, k class sparse representation method, can capture the same kind of sample feature information, strengthen the use of structure information, and can have the effect of primary, improve operational efficiency, in artificial data set, multiple features handwritten data set of a large number of experiments show that the method has good performance.