A Spatiotemporal Saliency Model of Visual Attention Based on Maximum Entropy

This paper presents a spatiotemporal saliency visual attention model based on maximum entropy. A dynamic saliency map is created by calculating entropies in every corresponding local region between the current and some previous frames in a short video. In the same time, four low-level visual features including color contrast, intensity contrast, orientation and texture, are extracted and are combined into a static saliency map for the last frame in the video. Our proposed model decides salient regions based on a final saliency map which is generated by integration of the dynamic saliency map and the static saliency map. At last, the size of each salient region is obtained by maximizing entropy of the final saliency map. The key contribution of this paper is that the entropy value method is used to calculate dynamic saliency for some successive frames in a short video. Experimental results indicate that: when the moving objects do not belong to the salient regions, our proposed model is excellent to Ban’s model.