An improved computational model for image saliency detection

This paper presents a new algorithm to solve the problem of saliency detection based on frequency-tuned algorithm, sparse code and combination strategy. Firstly, four broadly-tuned color channels are built for the input image. Secondly, saliency maps in every channel are computed using the frequency-turn algorithm. At the same time, another saliency map is computed for the original image using the sparse code model. Finally, we propose a new feature combination method to obtain the final saliency maps. Compared with previous methods, our saliency measure is more effective and robust as demonstrated by the experiments.