Skin-color based particle filtering for human face tracking

Skin color is a very important feature for the real time face tracking. By analyzing the skin color distributions of different people, we propose a novel face tracking algorithm which integrates the chromatic color information of the face skin region into the particle filtering (PF) framework. With the assumption of the Gaussian distribution of the face chromatic color, a Gaussian model is used to project the chromatic information of the YC_bC_r face image into a chromatic probability gray image. The histogram of the chromatic probability gray image is considered as the observation model for the PF. The update of the weight vector of the PF is determined by the Bhattacharyya distance between the reference model and the measured observed model. Extensive experiments have showed that our proposed algorithm performs quite well under the varying illumination, the full occlusion with the complex video background in terms of the tracking ability.