An improved algorithm for removing impulse noise based on long-range correlation in an image

An algorithm for removing impulse noise is proposed. The algorithm not only uses the information of neighboring pixels in a local region, but also uses the long-range correlation within a natural image. In the proposed algorithm, a search criterion based on the weighted transformed contents of regions is used to look for a region which is highly correlated to the region of interest and then the center of that region is selected to replace the corrupted pixel. The method is found to be very effective in removing impulse noise from corrupted images, both in terms of the objective distortion measure and subjective visual assessment.