Skin Segmentation Based on Graph Cuts  

作　　者：胡芝兰 王贵锦 林行刚 严洪 

机构地区：[1]Department of Electronic Engineering, Tsinghua University [2]Department of Electronic Engineering, City University of Hong Kong [3]School of Electrical and Information Engineering, The University of Sydney

出　　处：《Tsinghua Science and Technology》2009年第4期478-486,共9页清华大学学报（自然科学版（英文版）

基　　金：Supported by the National Natural Science Foundation of China (No. 60472028);the Specialized Research Fund for the Doctoral Program of Higher Education of MOE, China (No. 20040003015)

摘　　要：Skin segmentation is widely used in many computer vision tasks to improve automated visualiza- tion. This paper presents a graph cuts algorithm to segment arbitrary skin regions from images. The detected face is used to determine the foreground skin seeds and the background non-skin seeds with the color probability distributions for the foreground represented by a single Gaussian model and for the background by a Gaussian mixture model. The probability distribution of the image is used for noise suppression to alle- viate the influence of the background regions having skin-like colors. Finally, the skin is segmented by graph cuts, with the regional parameter y optimally selected to adapt to different images. Tests of the algorithm on many real world photographs show that the scheme accurately segments skin regions and is robust against illumination variations, individual skin variations, and cluttered backgrounds.Skin segmentation is widely used in many computer vision tasks to improve automated visualiza- tion. This paper presents a graph cuts algorithm to segment arbitrary skin regions from images. The detected face is used to determine the foreground skin seeds and the background non-skin seeds with the color probability distributions for the foreground represented by a single Gaussian model and for the background by a Gaussian mixture model. The probability distribution of the image is used for noise suppression to alle- viate the influence of the background regions having skin-like colors. Finally, the skin is segmented by graph cuts, with the regional parameter y optimally selected to adapt to different images. Tests of the algorithm on many real world photographs show that the scheme accurately segments skin regions and is robust against illumination variations, individual skin variations, and cluttered backgrounds.

关 键 词：graph cuts skin segmentation Gaussian mixture model （GMM） noise suppression 

分 类 号：R318.0[医药卫生—生物医学工程] TP391.41[医药卫生—基础医学]