A novel facial expression recognition approach based on MoSift

Obtaining compact and discriminative facial features is a difficult problem in expression recognition. In this paper, a novel approach to extract relevant features based on Mo Sift is proposed, which does not need advance detection of facial motion unit for expression classification. In the proposed approach, Mo Sift is adopted to extract interest points, and histogram of oriented gradient (HOG) and histogram of oriented flow (HOF) are adopted as feature descriptors. To get the discriminative representation, all expression features are clustered with the kmeans algorithm to learn a visual codebook. Then, unlike the traditional bag of feature (BoF) models using vector quantization (VQ) to map each feature into a certain code word, a sparse coding method named simulation orthogonal matching pursuit (SOMP) is applied. The proposed approach has been evaluated on the Extended Cohn-Kanade (CK+) database and the proposed approach has achieved an average of 86.35% accuracy.