| Image segmentation is one of the difficulties in computer vision which aims to partition animage into several constituents according to the properties and extract the region of interests. Thereare numbers of segment approaches in which graph cuts based interactive segmentation method isone of the best. It turned segment problem into the minimization of the energy function, and theresult of segmentation is obtained through solving the min cut in the flow network. Recently,android system has occupied most of the intelligent terminal market with its excellent userexperience, and the improvement of user demand for image sharing makes the image editingapplication in android platform more and more popular. In this paper, the graph cuts algorithm istransplanted into android platform, and also, a cutout android application is developed based on thealgorithm. The main work is as follows:(1) The regularizing parameter of the energy function in the novel graph cuts methods stronglyaffects the result of the segmentation. In this paper, a modified energy function is proposed based onthe novel one with the regional term removed and the boundary term enhanced. In addition, thepresented algorithm is implemented and tested in android platform，and the experimental resultsshow that it gets better segmentation results with the running time reduced.(2) GrabCut is an interactive foreground extraction method using iterated graph cuts. It reducesthe user interaction and introduces the Gaussian Mixture Model which is considered to be one of thebest segmentation approaches. The algorithm is also implemented in android platform where theimpact on segmentation accuracy of iterations is tested. The experimental results demonstrate thatthe most cost-effective segment result appears when the number of iterations is set to2.(3) One android cutout application is developed in this paper based on the GrabCut algorithmwith correction operation added in order to ensure the accuracy of the segmentation. And theextracted foreground can be pasted into other images like the PhotoShop does. The experimentalresults conform that the application performs well and has its practical value. |
