Towards exploiting change blindness for image processing

Abstract

Change blindness is a type of visual masking which affects our ability to notice changes introduced in visual stimuli (e.g. change in the colour or position of an object). In this paper, we propose to use it as a means to identify image attributes that are less important than others. We propose a model of visual awareness based on low-level saliency detection and image inpainting, which identifies textured regions within images that are the most prone to change blindness. Results from a user study demonstrate that our model can generate alternative versions of natural scenes which, while noticeably different, have the same visual quality as the original. We show an example of practical application in image compression.