Adaptive Classifier for Robust Detection of Signing Articulators Based on Skin Colour

The proposed classifier is a novel skin detector that outperforms most of the existing approaches by dropping most of the non-skin pixels in its earlier stages of weak classifiers. Only the pixels with maximum skin likelihood are processed in later adaptive classifier. Parametric background modelling and validation based online training significantly improves the robustness of the whole classifier in any daily-life lighting conditions.