Abstract:

A facial expression recognition method based on adaptive dictionary learning and sparse representation was proposed. Eyes and mouth were located by Haar-like features and the geometric structure of face. For learning dictionaries, random blocks near facial salient points were extracted, and those blocks were trained from the same kind of expression to form one expression dictionary. Then, all expression dictionaries were arranged into a learned dictionary. The test expressions were classified by sparse decomposition. The good performance was verified by the experiments on standard JAFFE and Cohn-kanade expression database. Good robustness was obtained even if the image had noise corruption or under occlusion.

Key words: adaptive, sparse representation, random block, facial expression recognition, dictionary learning