Human Skin Color Region Segmentation Based on Adaptive Model

We address the problem of segmenting human skin color region automatically in real world images,focusing on improving the robustness and accuracy of the segmentation result.We propose a novel method for skin color segmentation based on adaptive model.Firstly,AdaBoost face detection has been used to detect face area which containing light and skin color information in particular environment.The correction of the face area is regarded as a sample set of skin color.Secondly,Gaussian mixture model（GMM）,constructed by Y component and CbCr components in YCbCr color space,is selected as skin color model.Then,the similarity is measured by using improved Mahalanobis distance between image pixels with the skin color models.Finally,segmentation threshold is determined.The experiments demonstrate that our method can segment skin color region automatically and robustly.Compared to existed fusion approach and CbCr threshold method,our method improves about 10% detection rate and reduces the negative detection rate simultaneously.