Optimal Reconstruction of Natural Images by Small Sets of Gabor Filters

Abstract

Images can be reconstructed after being filtered by a Gaussian and a few Gabor filters. Several search methods for the filter parameters for a (near) optimal reconstruction are examined. At first, the search is performed on a 1-D signal which satisfies the radial spectrum of the average of natural images. Later, these results are extended and applied to a 2-D example. It is shown that the resulting parameters are better than those of some existing models. © 1998