Facial Expression Recognition Based On ASM&SURF

Facial expression recognition(FER) is a valuable research area, which is based on face recognition and plays an important role in the emotion computation field. It has been used in many fields, such as: psychological research, medical services, online teaching, safe driving and other daily life. We can enhance the ability of human-computer interaction by studying facial expression recognition. It also can make us using the intelligent human-computer interaction system more precisely, and make people working and living more conveniently.This paper introduces the knowledge of background and the significance of studying about facial expression recognition, it also describes the current study of facial expression recognition. We review the active shape model in detailed: the building of traditional ASM and the fitting of feature point by ASM. We also focus on the extraction of facial expression feature and facial expression recognition. My work as follows:(1) To construct the ASM sub-local feature model based on LBP. Due to the accuracy of the feature point location can directly affect the expression recognition result, we propose a local feature model based on LBP instead the local gray level model, which can more fully use the texture feature of feature points.(2) To propose average angle of double normal as the fitting orientation. The traditional ASM feature point fitting orientation is normal, which does not take full advantage of the geometric relationship of feature points, so the average angle of double normal is provided as the fitting direction.(3) To propose a facial expression recognition algorithm based on ASM&SURF integration feature and SVM. This paper gains the local texture feature based on Speed-up Robust Feature around feature points which is located by ASM. ASM&SURF integration feature is formed by the combination of ASM and SURF, which can both describe the shape feature and texture feature of expression.Finally, we use the ASM&SURF integration feature as SVM’s inputs to classify seven facial expressions. The experience is based on JAFFE and the average recognition rate of seven basic expressions achieves 93.125%. So the ASM&SURF integration feature can effectively represent the expression information.