Comparasion of graph cuts based interactive segmentation methods

In this thesis study, Interactive Graph Cuts, Grabcut and Lazy Snapping interactive image segmentation methods are compared. Graph cuts based methods consider each pixel of the image as a correspondence to a node and divide the graph into a foreground and background using the maximum flow-minimum cut algorithm to minimize energy function. By using the features obtained from the first-order statistical and Gray-level co-occurrence matrices (GLCM), the foreground and background images are divided into two groups, uniform and complex. These clusters are used in the comparison of graph cut based interactive image segmentation methods. In the segmentation results, it is seen that grabcut method is more successful than other methods in image cluster which the foreground is uniform. It has been determined that the Lazy Snapping method has better segmentation results than other methods in image clusters where the foreground is complex. It is also seen that that Lazy Snapping is the fastest method of performing segmentation in all image clusters. KEYWORDS: Interactive Image Segmentation, Graph Cuts, Grabcut, Lazy Snapping