Facial Expression Recognition Based on Block Gabor Wavelet Fusion Feature

Considering the complexity of facial expressions, this paper adopts the strategy of face segmentation for facial expression auto-recognition. On the training dataset, feature-based Gabor wavelets are extracted for each facial block. Then classification is done to realize the basic facial expression recognition to determine which part contributes most for a given expression. A binary weight matrix is assigned a weight of ‘1’ for highest such recognition rate, otherwise ’0’. At the test stage, the voted feature is calculated for each basic expression by multiplying the weight matrix and than the relative recognition rates are obtained. Finally the target expression is determined by voting. The experiment results show that Gabor feature plays better roles in representing the express than PCA and DCT. Based on the learning the optimal area wavelet modeling expressions, relative to the entire facial feature extraction, the system has better recognition effect.