Visual Attention Model with Cross-Layer Saliency Optimization

Detection of visually salient regions is useful for applications like image adaptation, adaptive compression, image retrieval and so on. In this paper, a new bottom-up visual attention model (VAM) is proposed based on the spirit of cross-layer optimization in the field of communication. In this model, the local saliency and global saliency are firstly extracted based on the contrast of low-level features from the local and global layers respectively, and then they are used to construct a weight model. Finally the proposed VAM is obtained by optimizing the global saliency with the weight model, which is taken as a feedback from the local layer to the global layer. Experimental results demonstrate that the proposed VAM performs competitively with four existing models on detecting out accurate salient regions and enhancing the contrast between salient and non-salient regions.