Study Of Issues Related To Kernel Sparse Representation

With the rapid development of multimedia and the Internet, the image data is growing at an alarming rate. How to effectively express and classify this mass storage of image becomes a hot research. In recent years, the emerging sparse representation theory as an image signal representation method, can deal with noisy images because of over-complete dictionary. Therefore, all fields of image processing have been widespread concern. In recent years, kernel trick are also used in the field of sparse representation. More and more sparse representation method based on kernel space and classification model based on kernel sparse representation is proposed.This paper looks at two different aspects of kernel sparse representation: algorithm for solving kernel sparse coefficient and classification model based on kernel sparse representation. On the one hand, the kernel orthogonal matching pursuit algorithm can solve kernel sparse coefficients, but its atomic orthogonal step has amount of computation. On the other hand, kernel label consistent K-SVD classification model considers the kernel sparse representation of the image, but not consider the kernel sparse representation of the identification coding.Focused on these problems, a fast kernel orthogonal matching pursuit was proposed. This paper analyzed its residual update rule and time complexity. Based on the algorithm, a double kernel label consistent K-SVD classification model was proposed. Experimental results show that, for the same kernel sparse representation problem, fast kernel orthogonal matching pursuit algorithm has faster speed than kernel orthogonal matching pursuit algorithm; double kernel label consistent K-SVD classification model has a higher classification accuracy and faster classification speed than the original classification model.