Multi-Cue-Based Face and Facial Feature Detection on Video Segments

An approach is presented to detect faces and facialfeatures on a video segment based on multi-cues, including gray-leveldistribution, color, motion, templates, algebraic features and so on.Faces are first detected across the frames by using color segmentation,template matching and artificial neural network. A PCA-based (PrincipalComponent Analysis) feature detector for still images is then usedto detect facial features on each single frame until the resultingfeatures of three adjacent frames, named as base frames, are consistentwith each other. The features of frames neighboring the baseframes are first detected by the still-image feature detector, thenverified and corrected according to the smoothness constraint and theplanar surface motion constraint. Experiments have been performed onvideo segments captured under different environments, and thepresented method is proved to be robust and accurate over variableposes, ages and illumination conditions.