Cross-modality automatic face model training from large video databasesXiaodan SongChing-Yung Linet al.2004CVPRW 2004