Kernel wiener filter using canonical correlation analysis framework

This paper addresses the problem of kernel Wiener filler using kernel canonical correlation analysis (CCA) framework. We solve the Wiener filter problem in the higher dimensional mapped domain using the kernel trick. A method is proposed to find approximate Wiener filtered signal in the original space by solving an optimization problem in higher dimensional space. The final form of kernel Wiener filter that relates to kernel Gram matrices, corresponds to the mean shift procedure or weighted nearest neighbor retrieval. The signal estimation and reconstruction capability of the kernel Wiener filter is demonstrated on the United States Postal Service (USPS) digits database. Moreover, a comparison between the linear Wiener filter and reduced-rank kernel Wiener filter is also presented