DETECTION OF IMAGE PAIRS USING CO-SALIENCY MODEL
In this paper a method is presented to identify co-attention objects from an image pair. This method provides an effective way to predict human fixations within multi-images, and robustly highlight co-salient regions. This method generates the SISM by computing three visual saliency maps within each image. For the MISM computation, a comultilayer graph is introduced using a spatial pyramid representation for the image pair. Two types of descriptors (i.e., color and texture visual descriptors) are extracted for each region node, which are then used to compute the similarity between a node-pair. Finally, a fast single-pair SimRank algorithm is employed to measure the similarity based on the normalized SimRank score.