Compact local Gabor directional number pattern for facial expression recognition

Abstract

This paper explores a novel method to represent face images for facial expression recognition; it is namedcompact local Gabor directional number pattern (CLGDNP). By convolving the face images with Gabor filters, we encodethe magnitude and phase response images in each scale, and calculate the histograms in several nonoverlapping regions ofeach encoded image. Finally, we obtain two spatial histogram sequences by the aid of the mean pooling technology andconcatenate them to form the facial descriptor. Moreover, for evaluating the performance of the proposed method, weemploy a support vector machine to conduct some extensive classification experiments on the Radboud faces database,the extended Cohn-Kanade database, and the Japanese Female Facial Expression database. The experimental resultsdemonstrate that the proposed CLGDNP method achieves better performance in classification.