Multiple Kernel Sparse Representation Based Classification

Hao Zheng,Fan Liu,Zhong Jin
DOI: https://doi.org/10.1007/978-3-642-33506-8_7
2012-01-01
Abstract:Sparse representation based classification (SRC) has been very successful in many pattern recognition problems. Recently, some extended kernel methods have been proposed through mapping the samples from original feature space into a high dimensional feature space, and then performing the SRC in the high dimensional feature space. However they are all simple kernel methods whose kernel is not most suitable one. For addressing this question, we proposed a novel method named multiple kernel sparse representation based classification (MKSRC), which combine several possible kernels and make full of kernel information. More importantly kernel weights of MKSRC can be automatically selected. The experimental results of face databases indicated recognition performance of new method is superior to other state-of-the-art methods.
What problem does this paper attempt to address?