Abstract:

Graph cuts is a combinatorial optimization method based on graph theory,and GrabCut based on it is an efficient foreground extraction algorithm.To achieve a certain segmentation accuracy,multiple iterations using graph cuts during the course of parameter estimation in the Gaussian Mixture Model(GMM),make it consume massive time when processing a great deal of image data.In this paper,we use quadtree decomposition to divide the image into several subblocks with internal high similarity and thus construct a compact weighted graph;the Gaussian Mixture Model (GMM) parameters can be estimated by the mean RGB values instead of all pixel values within blocks,so it can reduce the problem scale and significantly improve the efficiency of the algorithm.The experimental results demonstrate the feasibility and effectiveness of the algorithm.

Key words: graph cuts;quadtree decomposition;Gaussian mixture model