Interactive image resizing based on gradient intensity and color saliency

In this paper, we propose the novel interactive seam carving based on the importance map which is made by reflecting an intention of a user. First, a user provides only some foreground pixels as seeds for importance objects. Second, target object which the user aimed at is searched by the weighted histogram intersection based on gradient intensity and color. An importance map is defined based on gradient and color saliencies which were calculated from the obtained rectangle. At last, the optimal seam is searched and removed by using the dynamic programming. From resizing results using our seam carving, shape of target objects which the user intended at was preserved precisely. Also, unlike gradient maps, which may have to be recomputed several times during a seam carving based retargeting operation, our importance map is computed only once.