Study On Facial Feature Point Localization Based On Facial Expression Recogntion

The facial expression recognition is an essential aspect of the intelligent man-machine interface. It reflects the person’s pain and mental activities. Using the expression analysis to assess the psychological situation of passengers under the extreme operating conditions of trains is a new attempt. This paper aims to explore a real-time analysis scheme of passengers’expression information to evaluate the psychological condition of passengers. It provides first-hand feedback information for evaluating the degree of comfort of trains. In order to build the real-time facial expression recognition system, expression feature extraction using fast and accurate facial feature points localization is an effective method. This paper firstly introduces the commonly used method in facial expression recognition and feature point localization in recent years. It also discusses two common facial feature point localization algorithms, and then does the related improvement. Based on the feature point localization, feature extraction and expression classification are realized. Finally, the facial feature point localization and expression recognition system is designed.The major contributions of this paper are as follows:1. The two common facial feature point localization algorithms, namely, Active Shape Model (ASM) and Active Appearance Model (AAM) are introduced and realized. Advantages and disadvantages of the two algorithms are also analyzed.2. According to the shortcomings of the original two algorithms, two dual fitting methods are proposed. The one is Multi-Resolution AAM (MR-AAM) dual fitting method, and the other is the ASM-AAM dual fitting method. The two proposed methods can find the exact initial position and achieve better result of facial feature point localization.3. Through the facial feature points, the features of the facial expression are extracted. Then the feature extraction data is trained using support vector machine (SVM). The model of expression classification is obtained. It achieves the expression classification of the target image. The experiments are designed based on the JAFFE database. The experimental results show that the facial expression using the method of the feature point localization is effective.4. A facial feature point localization and facial expression recognition system is developed. It consists of such modules as feature points positioning, face detection, feature point localization and facial expression recognition.