Fast Computation Of Orthogonal Transforms Using Sparse Mixed Kernels

Abstract

A two-stage computation of orthogonal transforms that leads to faster algorithms with reduced numbers of multiplications is the subject of this paper. The first stage (a mixed transform) preprocesses the transform using a mixed transform matrix, while stage two is a correction which. in this case is given as a Walsh-Hadamard transform (WHT). Application to isotropic quadratic filters and multiplier-free implementation of orthogonal transforms are demonstrated.