关键词: 自适应模型, 肤色检测, YCbCr颜色空间, 期望最大化算法, 贝叶斯决策

Abstract: Because of the effects of environment, illumination and ethnicity, skin-color clustering in different images are not the same. As a result, for images with complex backgrounds, a fixed decision boundary skin-color model may lead to high false rejection rate and false detection rate. Based on the segmentation result of a fixed decision boundary skin-color model in YCbCr color space, a simplified Expectation Maximization (EM) algorithm was used to train the adaptive Gaussian model for a specific image. Combining the fixed model and the adaptive Gaussian model could produce the final skin color model for skin color segmentation. The experimental results show that the method can greatly improve skin-color segmentation accuracy, and reduce both false rejection rate and false detection rate.

Key words: adaptive model, skin color detection, YCbCr color space, Expectation Maximization (EM) algorithm, Bayesian decision