A novel support vector machine kernel based on Slepian semi-wavelets

In this paper, we construct a positive definite kernel associated with Slepian semi-wavelets. The kernel possesses multiscale structure and exhibits a strong localization property. It is convolution type associated with asymptotic sparse Gram matrix and allows the use of thresholding methods. We then focus on developing practical numerical algorithm to compute the kernel. Applications of the kernel in the context of kernel adaptive filtering are discussed.