Bottom-Up Saliency Estimation Based on Redundancy Reduction and Global Contrast

A new algorithm for bottom-up saliency estimation is proposed.Based on the sparse coding model,a power spectral filter is proposed to eliminate the second-order residual correlation,which suppresses the global repeated items effectively.In addition,aiming at modeling the mechanism of the human retina prior response to high-contrast stimuli,the effect of color context is considered.Experiments on the three publicly available databases and some psychophysical images show that the proposed model is comparable with the state-of-the-art saliency models,which not only highlights the salient objects in a complex environment but also pops up them uniformly.