A neural net model based on discrete Gabor transformation

It has been shown that Gabor representation can be effectively used for image analysis, segmentation and compression. A straightforward and efficient method is proposed for transforming discrete signals into generalized non-orthogonal Gabor representations. If both signal and the window function are real functions, complete Gabor coefficients can be found by multiplying a constant complex matrix and inverse of a sparse real matrix. A fast algorithm is suggested to compute the inverse of the matrix. Properties of Gabor coefficients based on the new method are discussed. A neural network model based on this method is proposed