Rank Constraint Based Co-saliency Detection

The traditional visual saliency detection method usually simulates the behavior of human visual processing systems to analyze a single scene quickly and automatically, which can be used to capture the most attractive regions in an image. The output of the saliency detection algorithm is called saliency map. The co-saliency model is the extension of visual saliency analysis to a group of input images, which aims at discovering the common salient objects existing in multiple images. Recently, as an effective preprocessing method, co-saliency enhances a host of computer vision and multimedia applications. Therefore, the proposed algorithm in this paper is focused on improving the performance of co-saliency detection.This paper first utilizes the intrinsic relationship of the co-salient objects to provide a rank constraint based fusion method for co-saliency detection. Given an arbitrary number of images with similar objects, our method employs several saliency detection algorithms to generate a group of saliency maps for these images. Based on the saliency maps corresponding to each image, our method segments a group of different salient objects, and uses feature histogram to describe them. The feature representation of the co-salient regions should be both similar and consistent. Therefore, the matrix jointing these feature histograms appears low rank. We formalize this general consistency criterion as the rank constraint, and propose low rank energy to describe it, which is based on low rank matrix recovery. In addition, the paper provides a reconstruction-based method to describe the rank constraint from an alternative view. In conclusion, two co-saliency detection models are proposed in this paper, and the first one is also valid for single image saliency detection. Experimental results on three benchmark datasets demonstrate that the proposed methods outperform the state-of-the-art methods.