Multi-scale stochastic color texture models for skin region segmentation and gesture detection

This work presents a novel method for skin detection as a pre-processing step for (hand) gesture segmentation. First, the skin color and texture models are learnt from a training set of skin images, where a Gaussian Mixture Model (GMM) and texton dictionary is constructed. Then, a stochastic region merging strategy is used to segment the image texture regions, from which each segment is classified based on the skin color and skin texture models. Compared with other state-of-the-art skin segmentation techniques, our experimental results suggest that our approach can handle better color and illumination variations arising from skin tone and pose changes, while keeping its ability for discriminating skin from other highly textured background materials.