Salient Object Detection Based on Commute-Time Distance

Aiming at the problem of saliency detection, this paper proposes a commute-time distance based salient object detection model to extract important objects of images. First, initial background prior map is generated based on background seeds at image boundary detected by clustering method. Next, salient points are utilized to get a coarse convex hull of salient region, the improved convex hull prior map is obtained based on the saliency seeds extracted from the convex hull. Finally, integrate the two prior maps to get the final saliency map. A novel distance named commute-time distance is employed to measure the feature difference between regions. Experimental results show the superiority of commute-time distance over traditional Euclidean distance or geodesic distance. Furthermore, effectiveness of the proposed model over many state-of-the-art methods is illustrated.