Accuracy Improvement of Graph-Cut Image Segmentation by Using Watershed
Traditional Graph-Cut algorithm traverses all pixels at each time of computation; consequently, it consumes a lot of time. This paper improves on Graph-Cut algorithm based on characteristics of Watershed. The basic theory is to insert watershed into Graph-Cut to conduct pre-segmentation on image. With watershed, image is divided into regions which have different sizes and pixel color similarities. Images processed by watershed algorithm are converted into weighted undirected graph; and then translate energy function on pixel into that graph on separate regions after pre-segmentation. Performance of test programs has proved that the improved Graph-Cut algorithm can increase workload of user interaction mark effectively. As long as workload considered in the interaction process, improved Graph-Cut algorithm can achieve ideal segmentation effect even on complex background.
J. Rong and Y. L. Pan, "Accuracy Improvement of Graph-Cut Image Segmentation by Using Watershed", Advanced Materials Research, Vols. 341-342, pp. 546-549, 2012