Facial Expression Recognition Using A CMAC Network with Clustering Memory

Abstract

In this paper, a facial expression recognition system based on CMAC (Cerebellar Model Articulation Controller) with a clustering memory is presented. Firstly, the facial expression features were automatically extracted and preprocessed from given still images in the JAFFE database in which the frontal view of faces were contained. A 2D DCT was then used to focus the key information of expression characteristics in order to decrease the size of images. Thirdly, a block size of the lower frequency of DCT coefficients is rearranged as input vectors with binary manner to send into the proposed CMAC that can rapidly obtain output using non-linear mapping with a look-up table in training or recognizing phase. Finally, the experimental results demonstrated recognition rates with various block size of coefficients in lower frequency and learning rates to show promising recognition results.