This paper presents a new and robustapproach to the detection, localization and tracking ofa human face in image sequences. First, a fast algo-rithm based on the neighbor-point-reliability is pro-posed to calculate the optical flow, which is used toextract the motion region. Then the hair and thehead knowledges are used to locate the face area. Forface tracking, a new genetic algorithms-based dynamictemplate-matching method is applied to search thenew position of the face in each new video frame. Ex-perimental results show that the proposed face track-ing method is fast and robust to illumination, faceposes, facial expressions and image distractors suchas facial occlusion by hands.