Research On Facial Expression Recognition Based On 2D Gabor Transforms And SVM

In recent years, with the increase of people's interest in the human-computer interaction, affective computing is becoming a research focus. Facial expression which carries rich informations of human behavior is the leading carrier of human affective. The research of facial expression recognition (FER) has important practical significances for the enhancement of computer's intelligence and humanity, the development of new human-computer environment and the promotion of disciplines such as psychology, etc. Eventually it will have major economic and social benefits.The main purpose of this paper is to recognize human facial expression. The research in this paper focused on the extraction and analysis of the facial features. This paper presents a new FER algorithm based on 2D Gabor transforms and SVM.The main innovations are as follows:(1) Locating facial expression region based on gray-level integration projection. Facial expression mainly in the eyes, eyebrows and mouth areas, known as the facial expression region. In this paper, automatic locating expression region using binary image and gray-level integration projection combination.(2) The main class of linearly-responding cortical neurons (called simple cells) are best modeled as a family of self-similar 2D Gabor wavelets.In this paper, a family of 2D Gabor wavelets transform the facial image, then Gabor features of each grid together form the feature vector of FER by the gridable facial expression region.(3) The facial features are classified using C-SVC. The multiclass C-SVC which is constructed by combining several binary classifiers is used for facial features classification. Automatic select parameter by combining grid-search and cross-validation, and a method is proposed for the unbalanced sample data. Furthermore, in this paper, through dropping the redundant features using sequential backward selection (SBS), improve the accuracy and speed of algorithm.(4) Based on the above researches, a facial expression recognition system is built. The system not only could test FER algorithm presented in this paper, but also could be used as experimental flat for other facial expression research.