An Action Unit Based Hierarchical Random Forest Model To Facial Expression Recognition

Facial expression can deliver rich emotional information,which not only increase the expressive effect of people,but also help people to understand the expression more accurately by the others.By feature extraction and machine learning,a facial expression recognition system is a computer application capable of identifying or verifying a person's emotion from a digital image or a video frame from a video source.Facial expression recognition plays an important role in the analysis of human emotion and the construction of harmonious human-computer interaction environment.In recent years,the research of facial expression recognition has made great progress,but the expression recognition in noisy environment is still a difficult problem.By summarizing the experiences of the previous research,this thesis is dedicated to improve the expression feature extraction method and the facial expression recognition algorithm:(1)This paper summarizes the research background and development prospects of facial expression recognition technology,and introduces three key steps of facial expression recognition including face detection and key area recognition,facial expression feature extraction and facial expression classification.This paper analyzes the advantages and disadvantages of the existing methods about the three key steps,which provides theoretical basis for the research.(2)To detect face and locate the key area,Adaboost based on haar-like feature is used.(3)For feature extraction,this paper proposes a method based on the facial action unit(AU)area.By the introduction the concept of facial AU area,the extracted features can accurately describe the subtle expression changes,and improve facial expression recognition rate.Furthermore,the proposed method can overcome the difficulty of accurate detection of faces AU itself,especially in low resolution and noise environment.(4)For facial expression recognition,this paper proposes a hierarchical random forests based on different AU,which can classify the facial expression layer by layer.The proposed method is compared with two other methods,SVM and traditional random forests,on the public dataset.The experiment results show that the method we propose has better result than SVM and traditional RF both in the noise environment and in the noise free environment,and it has better robustness in noise environment.