Video saliency detection based on mutual information and background prior in compressed domain

Saliency detection has been extensively studied due to its promising contributions for various computer vision applications. In this paper, we propose a novel algorithm to detect visual saliency from compressed video signals by exploiting mutual information and background prior. First, the local center-surround contrast principle is applied to detect the foreground saliency map. Separate spatial and temporal saliency maps are generated, where the computation of these saliency maps incorporates the mutual information of feature distribution between center window and surrounding window, and motion vector and the discrete cosine transformation coefficients including luminance, color and texture are used as the essential features. Second, before combining the spatial and temporal saliency maps, the background prior and selective background features from four boundaries of the video frame are implemented to suppress the background noises. To validate the effectiveness of the method, tests are carried out on a public database and experimental results show that the proposed method outperforms the existing state-of-the-art saliency detection methods.