Stochastic Context-Aware Saliency Detection

Most image retargeting algorithms rely heavily on valid saliency map detection to proceed. But the inefficiency of high quality saliency map detection severely restricts applications of these image retargeting methods. In this paper, we describe a stochastic algorithm for efficient context-aware saliency map detection. Our method is a multiple level saliency map detection algorithm which integrates multiple level coarse saliency maps into result saliency map and selectively updates unreliable regions of saliency map to refine detection results. With the virtue of randomized search, our method just needs very little extra memory beyond the input image and result map, and does not need build auxiliary data structures to accelerate saliency map detection. We implemented our algorithm on GPU, the performance of the proposed algorithm was demonstrated on a variety of images and video sequences, and was compared with the state of the art in image processing.