Research Of Image Pretreatment And Feature Extraction In Facial Expression Recognition

Facial expression recognition is a research topic that has theory foundation and application value. A typical facial expression system is usually including: face detection, image pretreatment, feature extraction, emotion recognition, and so on. The theory research of image pretreatment and feature extraction has significant importance in facial expression recognition systems. In this paper, we compared the related methods of image pretreatment and feature extraction by experiment. Furthermore, we design the approach of the image pretreatment and feature extraction for automatic facial expression system.In the research of image pretreatment, we discussed mainly the methods of illumination compensation and image size normalization. In the algorithms of illumination compensation, we discussed histogram gray transformation, linear transformation, exponential transformation and log transformation et al. According to the principle of real-time and accuracy, we use the histogram equalization to implement the system for the usual condition. In the algorithm of size normalization, eye location is an important and difficult problem. First we obtained the eye's location, and then converted face images into the normalized images by using size and angle transformation according to the transcendent principle. Except for this, we discussed various eye location methods and proposed an algorithm of eye location based on real-time for implementing the system.In the research of feature extraction, we discuss the model-matching algorithm in the feature extraction and propose a different model-matching algorithm. Its performance is as good as the wavelet algorithm, but has little operation quantity. For this it is appropriate used for simple real-time automatic recognition systems. Previous facial expression recognition usually implements the location by hand, but it is not fit for the automatic recognition system. Hence we discussed Active Shape Model (ASM)/Active Appearance Model (AAM), compared their advantages and disadvantages respectively, and then uses them for the expression recognition system.