Feature Extraction Algorithms Research Of Facial Expression Recognition

Artificial emotion is a new and attractive area in the intelligent control community. As the robotics mature, the fields of medical treatment, household, sports, services and so on are making increasing use of robots. They need the robots have the characteristics of intelligence, autonomy and interaction. Artificial emotion makes it possibility. Emotion recognition is an important field in artificial emotion research. When the robots can distinguish the humans' emotion, they can communicate with people and realize emotional intelligent.Human emotion recognition features include voice, gesture, facial expression, psychology signal (myoelectric signal, blood pressure, respiratory capacity, etc) and so on. Facial expression recognition is an effective and realizable feature in them.The algorithms of facial expression recognition system mainly contain images' preprocessing algorithms, feature extraction algorithms and classification algorithms. This paper mainly discusses the algorithms of feature extraction.There are many methods applied in facial expression recognition. Principal Component Analysis (PCA), Independent Component Analysis (ICA), Fisher's Linear Discriminants (FLD) and Local Feature Analysis (LFA) are some of the classic methods. Recently, many novel methods are applied in the facial expression recognition such as Artificial Neural Networks, Support Vector Machines, Wavelet Analysis, Hide Markov Model and Optical Flow, etc. Some of them can get high recognition rates, but the recognition processes are very complex.The purpose of the paper is to research and improve feature extraction algorithms to improve the correct recognition rate.This paper proposes two high performance facial expression recognition methods. One is based on discrete wavelet transform (DWT) and FLD, the other is based on wavelet energy feature (WEF) and FLD. In the method of DWT+FLD, wavelet transform is first employed to preprocess the original samples and resaved parts of the coefficients. Fisher's Linear Discriminants (FLD) is then used to feature Extraction. To get better recognition rates, we not only resave the low frequency but also some of the high frequency. Finally, the Nearest-Neighbor rule is used to classify the seven expressions (Anger,Disgust,Fear,Happiness, Normal, Sadness, Surprise) of the JAFFE database. Experiments show that the method is characterized by good recognition and reduced recognition time.The visual properties of the face, regarding the information about the facial expression, can be made clear by describing the movements of points belonging to the facial features (eyebrows, eyes, and mouth) and then by analyzing the relationships between those movements. Seen from the angle of image processing, the movements can be described by the changes of images' edge. The high frequency gained from wavelet transform can reflect the images' texture. In the method of WEF+FLD, we construct wavelet energy feature (WEF) by the high frequency to describe the facial images' texture and use it to describe the facial expression. Besides, it has been confirmed that FLD is a very valid method. It's critical that it can describe the details of the image, which just meets the demand of facial expression recognition. So we combine FLD with WEF. The experiments show that the method can get high recognition rates and is easy to realize.