Facial Expression Recognition Based On AdaBoost And SVM

Facial expression is the most important vector of human emotion, which contains a wealth of psychological and emotional information, and facial expression is the most important non-verbal communication methods, with the rapid development of computer technology and human-computer interaction theory of continuous improvement, facial expression recognition has become research focus. Facial expression recognition is a cross-subject which includes image processing, pattern recognition, psychology and other disciplines. The research of facial expression recognition (FER) has important practical significances for computer humane, intelligent premise and analysis of human emotions.This paper based on processing of facial expression recognition, focus on the key issues of the facial expression recognition in image preprocessing, feature extraction and facial expression recognition, then build a facial expression recognition system.The main contents of this dissertation are as follows:1. The human face detection algorithm is studied, especially the algorithm which based on the class of Haar features and AdaBoost algorithm and with which achieved the eye detection successfully. Geometric pre-processing which include rotation, clipping and scaling, meanwhile lighting pre-processing such as histogram equalization are adopted.2. In this paper, a family of 2D Gabor wavelets transform are used to get the features. Determine the appropriate frequency scale and direction by the analysis of rate and real-time of facial recognition. With the method that combines AdaBoost with PCA (Principal Component Analysis) to reduce the expression feature vectors. Then the kernel function and parameters of SVM are optimally selected through the classification performance experiment. Finally , a SVM-based facial expression classifier is designed by comprehensively considering the recognition rate and the computational efficiency. 3. An automatic facial expression recognition system is built which based on the research in this paper. Through the specific face and face non-specific experimental test for the system, not only verify the validity of the algorithm also demonstrated that the system can be used as research and experimentation platform.