Combining motion and appearance for gender classification from video sequences

We investigate whether combining appearance (face structure) and motion (the way a person is talking and moving his/her facial features) boosts gender classification from face sequences. We propose and compare different schemes based on appearance only, motion only, and combination of appearance and motion. Experiments on various face video datasets of persons uttering phrases or expressing emotions show that combination of motion and appearance is useful for gender analysis of familiar faces, yielding in classification accuracy of 100%. However, for unfamiliar faces, motion seems to not provide additional discriminative information as the best performance (96.3%) is obtained using an appearance based approach with Local Binary Pattern (LBP) features and Support Vector Machines (SVMs).