Separable approximation methods for orientational filters and VLSI implementations

摘要

Abstract: Two separable approximation methods, Chebyshevapproximation and SVD approximation, are described asways of implementing an orientational filteringoperation on a 2D image. They are both suitable forparallel implementation and enable the orientationalfilter to be performed on an N $MUL N image inO(N$+2$/) time with a small amount of hardware. It isshown that these algorithms can be combined withShensa's algorithm so that a multi-resolution pyramidcan be efficiently constructed from orientationalfilter wavelets. Performance of the algorithms isevaluated by applying them to both Gabor filters andthe first derivative of Gaussian filters. A VLSIarchitecture for orientational filters using theseparable approximation is presented. !4