Saliency detection based on scale selectivity of human visual system

Abstract

It is well known that visual attention and saliency mechanisms play an important role in human visual perception. This paper proposes a novel bottom-up saliency mechanism through scale space analysis. The research on human perception had shown that our ability to perceive a visual scene with different scales is described with the Contrast-Sensitivity Function (CSF). Motivated by this observation, we model the saliency as weighted average of the multi-scale analysis of the visual scene, where the weights of the middle spatial frequency bands are larger than others, following the CSF. This method is tested on natural images. The experimental results show that this approach is able to quickly extract salient regions which are consistent with human visual perception, both qualitatively and quantitatively. © 2011 Springer-Verlag.