摘要
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C-支持向量机的方法进行图像的肤色检测。在YCbCr颜色空间,去除照度分量,用像素点及其8邻域内各点的Cb、Cr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C-支持向量机肤色检测器。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0.977。
According to the theory of Support Vector Machine and the characteristic of complexion distribution, a novel complexion detection approach based on C-Support Vector Machine and with pixels' 8 neighbours was proposed. In YCbCr colour spaces, taking Cb and Cr sub-vectors of one pixel excluding illuminance sub-vector and all Cb and Cr sub-vectors of its 8-neighbours' pixels to form the input vector, pixels' categories as output, Gaussian function as kernel function, using sequence minimal optimization as learning algorithm, complexion detectors based on C-Support Vector Machine were constructed. Experimental result shows that the correctness rate of the complexion detector reaches 0. 977 when the kernel width be 80 and the punish coefficient C be 200.
出处
《计算机应用与软件》
CSCD
2009年第5期202-204,共3页
Computer Applications and Software
关键词
肤色检测
支持向量机
颜色空间
高斯函数
8邻域点
Skin detection Support. vector machine Colour space Gaussian function 8-neighbours