SPACE AND FREQUENCY VARIANT IMAGE-ENHANCEMENT BASED ON A GABOR REPRESENTATION

In this paper we present an image enhancement method to remove degradations due to large amounts of space and frequency variant fractal noise. Images degraded by fog are examples of such a situation. First, a pyramidal image representation has been applied, consisting of 4 X 4 gaborian channels in frequency and orientation plus a zero-frequency residual. Previously, a set of channels corresponding to a generic undegraded image prototype was computed. By using that prototype along with a generic mathematical model of the noise (i.e., brownian fractal) it is possible to perform a blind image enhancement. The enhancement process is recursively applied at the different pyramid levels. The method has been tested by merging 90% of a fractal cloud cover with 10% of an aerial image, illustrating how many details practically invisible in the foggy image become clearly visible.