Video Salient Object Segmentation Based On Superpixel And Graph Cut

As video object segmentation is one of the basic research topics in the field of computervision, which is of great importance. video salient object segmentation focus on segmentinginteresting foreground object from video frames, which has a wide range of applications,such as: video summarization, human behavior recognition, video retrieval, video objectrecognition, video analysis and so on. This paper represents a video salient objectextracting method on superpixel (i.e. image over segmentation) level, which is based ongraph cut.The major work in this paper is as follows: at first, having surveyed visual saliencydetection, visual attention models and video object segmentation algorithm; secondly,having proposed a novel method for video salient object segmentation based on superpixeland graph cut; finally, having done the experiments on videos chosen from standard databaseand analyzed and compared the results.In this paper, I proposed a novel method for video salient object segmentation based onsuperpixel and graph cut which is the core work of this article. At first, the algorithmcalculates the image over-segmentation of each video frame. Secondly, the algorithm thenextracts static and dynamic characteristics of each video frame on superpixel level andcalculates the static saliency map and dynamic saliency map of each video frame. Thirdly,based on the principle of dynamic priority of human visual system, the algorithm fuses thestatic and dynamic map. Fourthly, the algorithm calculates objectness for each video frame.Fifthly, the algorithm builds the energy function as a combination of multiple measures suchas saliency, objectness and persistence. Finally, the algorithm uses iterative graph cut tominimize the energy equation, segmenting salient objects in each frame of a video sequence.The experimental results show the proposed method segmented video sequence iscompetitive.