The Research On HITS-based Image Saliency Detection

With the development of society, image saliency detection algorithms have applied into many fields, such as object recognition, image compression, video retrieval and so on. This paper propose a new image saliency detection model with the combination of three priors which we represent in a different way from traditional description and modified HITS algorithm.Our proposed algorithm is made up of three parts. First, based on an existing contour detection method BSR, we propose an adapt thresholding method to achieve image segment. Second, taking the above segment as basic processing unit, we compute initial saliency map from three priors. Our contrast is described with the rate of difference with adjacent segments and similarity with nonadjacent segments. Our location is described with image center, the shortest distance with the four image edges and the overlap with image boundaries. The last, we modified HITS algorithm to make the whole of object more uniform and partly correct detection errors.At last, based on the international open image database MSRA-1000, this paper compare ours with some the-art-of-state methods, the experiment result demonstrate the effective of our proposed method.