An improved technique of skin-color modelling for face detection in video is proposed. The skin-color is modelled as a combination of histogram based and data-mining based approaches. To solve a skin-color modelling problem in the presence of lighting variation, unconstrained environment, variation in races, resulting skin-color model is a product that minimizes registration errors and increases robustness and accuracy of the face detection in video system. The method improves applications performance on 7%, which is a significant advantage.