Detecting Facial Features on Image Sequences Using Cross-Verification Mechanism

Abstract

An approach is presented to detect facial features on image sequences. Face regions are first detected across the sequence, then a PCA-based feature detector for still images is used to detect facial features on each single frame until the resulted features of three adjacent frames donx’t change significantly. Given a frame with correct features, the features of its neighbor frames are first detected by the still-image feature detector, then verified and corrected by using the smoothness constraint and the planar surface motion constraint. Experiments have been performed on image sequences taken under different environments, and prove the presented method to be quite robust and efficient over variable poses, ages and illumination conditions.