A new co-saliency model via pairwise constraint graph matching
Fanman Meng, Hongliang Li, Guanghui Liu · 2012
In this paper, we propose a new co-saliency model to extract co-saliency maps from a pair of images. Rather than using unary constraint matching, we use pairwise constraint graph matching to obtain more accurate co-saliency map. In our method, the co-saliency map consists of two terms, i.e., the single image saliency map and the multiple image saliency map. The single image saliency map is obtained by traditional saliency detection method. The multiple image saliency map is extracted by matching the similar regions among the images, which is casted as pairwise constraint graph matching problem. The dynamic programming method is used to solve the matching problem. We test the proposed co-saliency model on co-saliency dataset. The experimental results demonstrate the effectiveness of the proposed co-saliency model.