Proto-Object Based Saliency Model with Second-Order Texture Feature

The nervous system can rapidly select important information from a visual scene and pay attention to it. Bottom-up saliency models use low-level features such as intensity, color, and orientation to generate a saliency map that predicts human fixations. Such algorithms work well for many images, however they miss the influence of texture. In this paper, we add a second-order texture channel to a proto-object based saliency model. The extended model shows significantly improved performance in predicting human fixations.