Skin segmentation using Possibilistic Fuzzy C-means clustering in presence of skin-colored background

Skin color segmentation is an important step in vision based Human Computer Interaction (HCI). But, accuracy of the color-based skin detection methods are severely affected by the presence of skin-like colors in the background. In this paper, a skin segmentation method for tackling a specific case of foreground and the background color similarity is proposed. An initial skin mask is obtained by a dynamic thresholding of the skin probability map. The Possibilistic Fuzzy C-Means (PFCM) clustering is employed to group the initially detected pixels into two clusters: a true-skin cluster and a false-skin cluster. Experimental results show that the proposed method can perform well when there is a color similarity between the skin-colored foreground regions and the scene background.