Research Of Methods For Facial Expression Recognition

Facial Expression Recognition(FER) has been the hotspot of research in pattern recognition field all over the world since it can be widely applied on many aspects in the future. As a subset of Facial Expression Recognition, smile recognition is also a research hotspot. Following this topic, we research the key Algorithms for smile recognition and multi-Expression Recognition in this paper. A facial expression recognition system contains face detection, face feature extraction, feature selection and expression classification. In these parts, feature extraction has great influence in recognition system.The key algorithms of facial expression recognition are studied in this paper and we focus our attention on the research of methods for feature extraction. As a kind of texture presentation feature, Gabor feature is a successful feature extraction method in facial expression recognition field. However, the dimension of Gabor feature is too large, and we usually need a long-term feature selection period before classification. What's more, the recognition rate can be further improved. In this situation, we introduce PHOG(Pyramid Histogram of Oriented Gradients) feature and BIM(Biologically Inspired Model) feature to facial expression recognition. The two feature extraction methods are widely used in classification recently years. Finally, a series of experiments are designed and implemented to compare different feature extraction and classification methods.The major works of this paper are listed as below:1. Introduce the research background and importance of facial expression recognition and the major methods for feature extraction and expression classification.2. Introduce the primary algorithms for face detection, in which the method based on Haar-like feature and Adaboost is introduced in detail, and an experiment is design to verify this method.3. Introduce PHOG feature and BIM feature to facial expression recognition field, and compared them with Gabor feature to verify their effectiveness.4. Use the method of cascading different feature and AdaBoost to get further improved feature representation, and obtain a more recognition rate.5. The classification methods SVM, AdaBoost and AdaBoost+SVM are used in our experiments, and their recognition performance are compared based on the results.The results of experiment verify the effectiveness of the feature extraction and classification methods we use in this paper.