Facial Expression Recognition Method Based On Feature Fusion

Facial Expression Recognition(FER) is a challenging issue in the field of multi-science and has tremendous research value and application prospect in Fatigue driving, Intensive Care,Animation and other fields. As a way of biometric identification, facial expression recognition is an indispensable part of human-computer interaction. Facial expression recognition research is to develop a high performance system to automatically identify human facial expressions and analyze their emotions. It is aiming at enhance the friendliness and intelligent of human-computer interaction.To the problem that Local Binary Pattern(LBP) is sensitive to noise and cannot describe texture's orientation information, a novel method by fusing Lifting Wavelet(LW) and LBP-Gradient Direction(LBP-GD) feature is proposed for facial expression recognition.LBP-GD feature which fusing LBP and Gradient Direction features, improves the traditional LBP operator, can effectively represent texture's orientation information and maintain the advantages of LBP itself. When extracting texture characteristics, an irregular partition way is proposed for the difference of information in different regions of human face. The image is partitioned into 9 regions with different weights and LBP-GD feature is extracted from each of region. At last, low frequency components and LBP-GD features are fused with weights,and the K-nearest neighbor method is used for classification. Experiments on JAFFE and Cohn-Kanade facial expression databases indicate the superior performance of the proposed approach.