A new algorithm for detecting co-saliency in multiple images through sparse coding representation

We propose what we believe to be a new algorithm for detecting the co-saliency in multiple images. First, we use the independent component analysis to learn and obtain a set of sparse bases of a natural image through filtering the input image and then use them to work out the sparse coding representation of the image to be detected. Second, we define the multi-variable Kullback-Leibler (K-L) divergence to measure the similarity among multiple images. Third, according to the properties of the K-L divergence, we detect the region where the divergence decreases significantly, or the similarity of the image, thus detecting the co-saliency in multiple images. To verify the effectiveness of our algorithm, we test the image co-saliency detection effect with the photos we took. The test results, given in Fig.3, and their analysis show preliminarily that the image co-saliency detection effect of our new algorithm is the same as that of human visual characteristics.