Abstract: Through the study on skin features under different light environments in YCgCr color space, it was found that the skin pixels have different Cg and Cr areas in different environments. The correlation matrix was used to estimate the light environment. The different skin segmentation methods were ultilized according to the result of environment detection, and dynamic threshold was composed by between-cluster variance and within-class scatter. Compared with the traditional methods, the proposed method fully reduces the impact of color distortion in different light. The experimental results show that it has higher accuracy and lower detection error rate.

Key words: YCgCr, correlation matrix, mixture Gauss model, automatic threshold

摘要： 通过在颜色空间YCgCr里对不同光照环境下的肤色特征进行研究,发现不同光照环境下的肤色像素高度聚集在不同的Cg、Cr区域。对此,提出了利用相关矩阵法对图片的光照环境进行估计,然后针对图片的光照环境检测结果采用不同的肤色分割方法,并采用一种基于类间方差和类内离散度相结合的动态阈值方法来确定分割阈值。该方法与传统的肤色检测方法相比,能够充分减少不同光照环境下带来的颜色失真的影响,是一种光照自适应的肤色检测方法。实验表明,该方法对肤色检测具有较高的准确率和较低的误检率。

关键词: YCgCr, 相关矩阵, 混合高斯模型, 动态阈值