Automatic detection and extraction of perceptually significant visual features

Perceptual-based algorithms attempt to discriminate between signal components based on their perceptual significance to the human receiver. This paper presents a simple and efficient algorithm for the suppression of nonessential visual features, while retaining those features that are important for the recognition of a scene by a human observer. The first step produces a perceptual mask, which is a spatial perceptual weighting map. This mask assigns perceptual significance to the different areas of the input image, and is used to derive an output image in which the non-essential features of the original image are suppressed. The presented algorithm is motivated by established psychovisual principles related to figure-ground perception and visual illusions, which show that the human visual system is capable of "filling in" missing details when presented with enough visual cues. Very good reconstructed images were obtained despite the reduction in information content. Examples are presented to illustrate the performance of the algorithm.