Abstract: This paper studies relativity between color and its parameters.Since color’s parameters are correlative，decreasing dimension by wiping off some parameters affects classificatory precision.Through normal approach，we find color in rg plane is similar to color in hs color，and they both have positive excursion in Q-Q plot.This excursion shows that decreasing dimension is improper and luminance must be considered in skin model.We propose a spherical skin model including luminance information in HLS color space.Then we compare this model to other models in our own database and PICS database.The model has successfully applied in these images.This model can be directly used in face detection for image and video.

Key words: relativity, normal approach, Q-Q plot, spherical skin model, face detection

摘要： 通过对颜色与其参数的相关性分析，认为颜色的各个参数都是紧密相关的，去除某些参数进行降维处理，实际上造成了精度损失。通过对rg颜色和hs颜色进行正态拟合，发现两者的QQ图显示出相似的正偏度，这说明降维是不成功的，亮度信息不能丢弃。在HLS空间提出了一个包含亮度信息的球形肤色模型，在自建库和PICS库上使用该模型，与其它模型进行比较，取得了比较好的效果。该模型可以直接用于图像和视频人脸检测。

关键词: 相关性, 正态拟合, QQ图, 球形肤色模型, 人脸检测