Chinese Journal of Computers   Full Text
Title	A Fast Object Extraction Method Based on Color and Texture Information
Authors	MU Ya-Dong ZHOU Bing-Feng
Address	(Institute of Computer Science and Technology, Peking University, Beijing 100871)
Year	2009
Issue	No.11(2252—2259)
Abstract &   Background	Abstract In recent years researchers have developed many algorithms for object extraction and image matting. However, previous approaches usually require trimaps as input, or consume intolerably long time to get the final results, and most of them just consider the color information. This paper proposes a novel fast hierarchical object extraction method. First the input image is segmented roughly into two regions: foreground and background, using a modified hierarchical Graph Cuts algorithm. After that, the opacity values for the pixels nearby the foreground/background border are estimated using belief propagation (BP). Unlike traditional BP-based approaches, besides the smoothness and color constraints, the texture information is introduced by building grayscale co-occurrence matrices. Moreover, considering the fact that the resolution of photographs taken by digital cameras continues to increase at a rapid and steady pace, the modified version of hierarchical Graph Cuts proposed in this paper could accelerate the above-mentioned computation process, getting a comparably satisfactory local optimal solution as previous approaches. Experiments show that the method is effective and efficient especially for large images. Keywords hierarchical graph cuts; belief propagation(BP); co-occurrence matrix; Markov ransom fields Background In the past few years, various image matting and object extraction methods are proposed, yet few of them are suitable for images with large size. We set out to find a tradeoff between accuracy and running time, which is obtained here via a modified version of graph cuts algorithm. Moreover, most of current approaches only use color information, which is insufficient for the difficult task of object extraction and image matting. In this paper, the authors attempt to incorporate more kinds of information besides color, by introducing co-occurrence matrix for the purpose of transferring texture information. This work is supported by the National Natural Science Foundations of China under grant No.60573149.