1、第 26卷第 5期null null null 计算机应用与软件 Volnull26No. 52009年 5月null null ComputerApplications and Software May 2009 _ # _ 徐从东null 陈null 春null 朱安国( b + a a 1 _ “ null g 230031)l : 2007- 11- 29b V , = , 5 : T M Y , m ) bK null 1 null null 根据支持向量机理论和肤色信息分布特点,提出利用像素点的 8邻域信息,用 Cnull支持向量机的方法进行图像的肤色检测b在 YCbCr颜色空间,
2、去除照度分量,用像素点及其 8邻域内各点的 CbaCr分量构成的向量作为输入,像素点所属类别为输出,高斯函数为核函数,采用序列最小最优化学习算法,构造了 Cnull支持向量机肤色检测器b实验表明,当核宽度为 80,惩罚系数 C为 200时,该肤色检测器的检测正确率可达到 0. 977b1 o M null null 肤色检测null 支持向量机null 颜色空间null 高斯函数null 8邻域点COMPLEXION DETECTION BASED ON SUPPORT VECTORMACHINEAND NEIGHBOURHOOD INFORMATIONXu Congdongnull Chen
3、 Chunnull ZhuAnguo(Departmentof Artillery Command Automation and Simulation, Artillery Academy ofPLA,Hefei230031,Anhui, China)Abstractnull null According to the theory of SupportVectorM achine and the characteristicof complexion distribution, a novel complexion detecnulltion approach based on CnullS
4、upportVectorM achine and with pixelsnull 8 neighbourswas proposed. In YCbCr colour spaces, taking Cb and Crsubnullvectors of one pixel excluding illum inance subnullvectorand allCb and Cr subnullvectors of its8nullneighboursnull pixels to form the input vector,pixelsnull categories as output, Gaussi
5、an function as kernel function, using sequenceminimal optimization as learning algorithm, complexiondetectors based on CnullSupportV ectorM achinewere constructed. Experimentalresult shows that thecorrectness rate of the complexion detectorreaches 0. 977 when the kernelwidth be 80 and the punish coe
6、fficient C be 200.Keywordsnull null Skin detectionnull Support vectormachinenull Colour spacenull Gaussian functionnull 8nullneighbours0null null _ m | 8 V b “ j m “ ! ! # , m u s t b 1 j $ “ d , _ / ,B _ Z E d 1 N b M , 4 _ Z E 1- 3, * 4$ : . _ / 8 J.9 , 2006, 29(2): 194- 207. 2 VezhnevetsV, Sazono
7、vV, AndreevaA. A Survey on PixelnullBased SkinColorDetection TechniquesC. Proceedings Graphiconnull2003, M osnullcow, Russia, 2003: 85- 92. 3 Phung S L, Bouzerdoum A, ChaiD. Skin segmentation using color andedge informationC. in Proc. Int. Symposium on Signal Processingand its Applications, 2003: 1- 4. 4 SimonHaykin. * M . : . # .v : v v ,1995. 3 Z , Z , . VisualC+ Y I L . : , 2004.
