Two Applications of Graph-Cuts to Image Processing

Graph-cut methods have become an intensive area of research in image processing, with applications to stereo vision, binary thresholding, classification and image de-noising amongst other areas. This paper describes novel graph-cut implementations for use in two applications: 2D optical flow and interactive object delineation. The optical flow implementation is based on recursive application of graph-cuts, and is shown to perform favourably with state-of-the-art methods. The delineation method is novel in that it requires only a single point to be selected for a delineation to be achieved, and uses a different type of graph topology compared to the traditional methods.