Exploiting Change Blindness for Image Compression

Abstract

Change blindness refers to the inability of our visual system to memorize details in pictures. We suggest that, under certain conditions, regions containing such details can be identified and altered in a way that benefits image compression. We define change blindness maps and a compliant texture synthesis method to identify, remove and subsequently recover pixels which are subject to change blindness. Unlike comparable methods, ours is particularly intended for images with complex textures. We demonstrate via a subjective experiment that up to 15% of such an image can be altered in a way that is considered as natural and acceptable.