Geodesic distance and compactness prior based salient region detection
Li Juan Zhou, Song Yang, Yaoming Yang, Zhaohui Yang · 2016
In this paper, by using geodesic distance and compactness prior, we present some effective improvements concerning the two important aspects of diffusion-based methods: the construction of the diffusion matrix and seed vector. First, based on the geodesic distance, we construct a 2-layer knn graph. Compared with the most frequently used 2-layer neighborhood graph, our graph does not only effectively use the local spatial relationship, but also removes the dissimilar redundant nodes. Second, we use the background weighted contrast and compactness prior to obtain the seed vector, compared with the previously most used boundary prior, our approach can better distinguish the saliency seeds from the background seeds. Experimental results on public benchmark image sets show the superiority of our proposed salient region detection method.