Research On Algorithm Of Facial Expression Recognition

Facial expression recognition is to analyze and detect the special expression state from given expression images or video frames and determine the subject's specific inborn emotion so as to achieve smarter and more natural inter-action between human beings and computers. Facial expression recognition has potential application values in many fields, including psychics study, image understanding, synthetically face cartoon, video retrieval, robot technology, virtual reality, and research and develop of new human-computer-interface environment based on facial expression.The system of facial expression recognition generally consists of face expression image capture, preprocessing, face detection and location, face segmentation and normalization, facial expression feature extraction, facial expression recognition. The task in this paper focuses on the key issues of facial expression recognition, such as feature extraction, feature selection and facial expression classification, and so on. The performances of proposed methods are illustrated by simulation experimental results.(1) Preprocessing of expression images is the first step in the whole recognition system. In this paper, we proposed a new algorithm based on the complexity and template match. It can detect directly and locate the human eyes in human face images without determining the human face position in advance. The experimental results show that this algorithm is simple and convenient. It establishes foundation for clipping and scaling, photometric preprocessing and histogram equalization.(2) The methods of the feature extraction based on Two-dimensional Gabor transform are discussed in details. Gabor wavelets in 5 scales, 6 orientations for extracting features are constructed because of Gabor coefficients'lower sensitivity to variations of lighting and position. Each image is filtered by 30 Gabor functions. After Gabor filtering, the amplitude values are used as facial expression features. The experimental results show that this method is better than PCA or 2D-PCA. On the other hand, Gabor coefficients have large dimensions, we propose two times sampling method to reduce dimension. The first one is irregular sampling which significantly reduces the dimension of feature vector and improves the recognition rate; the second one is using 2D-PCA method. Through the sub-space mapping, the most representative features are extracted to reduce dimension again. The experimental show that this method can reduce dimension obviously. (3) Nearest neighbor classifier, Euclidean distance classifier, Cosine distance classifier are used to classify seven expressions including angry, disgust, fear, happy, neutral, and sad and surprise. Then, fuzzy integral is applied to fuse outputs from results of three classifiers to get the final recognition result. The experimental results indicate that this improve method has higher recognition rate than method of based on average recognition rate for fuzzy density.