Approximation Kernel 2DPCA by Mixture of Vector and Matrix Representation

The computational complexity of kernel 2DPCA is mainly determined by the product of the number of input image samples and rows of matrix. For the large number of this product, it is computational intractable to directly calculate the eigenvalues and eigenvectors of kernel matrix in kernel 2DPCA. In this paper, a method called M2D-PCA is proposed to approximate the Kernel 2DPCA by mixture of vector and matrix representation. The original image matrix is divided into several image blocks, each image block is transformed into a vector and a new matrix is constructed by considering these block vectors as rows. By this way the complexity of proposed kernel M2D-PCA is decreased. It is shown by experiments that the performance of one of the proposed M2D-PCA is the best among the compared methods.