Salient object segmentation based on graph cut
Ran Shi, Zhi Liu, Yinzhu Xue · 2010
Salient object segmentation is an important technique for many content based applications. This paper presents an unsupervised salient object segmentation method under the graph cut optimization framework. First, we exploit a kernel density estimation based saliency model to generate the saliency map, which provides the useful cues for object segmentation. Then we exploit the saliency map to adaptively define the region cost term, the boundary cost term and their balancing weight in the cost function, which is minimized using graph cut to obtain a binary segmentation of salient objects. Experimental results on a variety of images demonstrate the better segmentation performance of our approach.