Salient Object Detection Based On Label Propagation

Since salient regions contain important information, salient object detection aims at finding interesting regions in images and videos. The detected regions can be further used for more complicated computer vision applications, such as object detection and recognition, image compression, content-based image editing, and image retrieval. The greatest challenge in salient object detection is to highlight salient objects uniformly and suppress the irrelevant background noise.In this paper, we propose a novel saliency detection algorithm based on label propagation. In order to achieve highlighting salient objects uniformly and suppress the background, several key techniques are exploited to detect salient objects, including SLIC algorithm, image prior acquisition, graph model construction and Bayesian integration. A key observation is that image saliency can be estimated by propagating the label information extracted from the background and object regions. Based on this idea, detecting salient object by label propagation consists of three basic steps. First, some boundary superpixels serve as the background labels and the saliency of other superpixels are determined by ranking their similarities to the boundary labels based on an inner propagation scheme. Second, we utilize objectness measure to pick out foreground nodes and propagate foreground labels based on the same inner propagation scheme. Third, we develop a saliency integration mechanism based on Bayes formula to combine the background and foreground saliency maps.Results on three benchmark datasets with accurate annotations show that the proposed method achieves comparable performance to the state-of-the-art methods. Compared with other algorithms, our algorithm could detect salient objects more accurately.