肤色检测最优空间
Optimum Colorspace for Skin-tone Detection
-
摘要: 在综述现存肤色检测算法所使用不同颜色空间的基础上,基于一个含有1 894张图像的样本库,采用肤色内聚性和肤色-非肤色类可分离性指标以及相应的分类器性能评估准则,考查了包含所有常用颜色空间的15个3维颜色空间和15个2维色度平面,建立了一个完善的肤色空间评价体系.结果表明:线性颜色空间是最优的肤色空间类,去除亮度信息将降低肤色-非肤色的类可分离性,并导致分类性能的退化.Abstract: In this paper,we compared 153D colorspaces and 152D chrominance planes widely used in skin-tone detection in terms of compactness of skin class,discriminability between skin and non-skin classes as well as the classification performance of several detectors based on a data set of 1894images.The results show that 1) the linear transform colospaces have better performance than others,and 2) the absence of the luminance component decreases performance significantly.
下载: