Research On The Technology Of Facial Expression Recognition Based On Feature Fusion

Recently, facial expression recognition has received the concern of more and more researchers, which is regarded as a kind of new human-computer interaction mode. This paper gives a comprehensive survey and analysis of the existing facial expression recognition technology and methods. By systematically analyzing and comparing of existing facial expression recognition methods, we present some novel methods such as localization of facial feature, feature extraction and expression recognition. The main work of this dissertation includes:(1) A mixed facial feature locating method based on linear object class of face and active shape model is presented. This method firstly reconstructs the facial image from the training samples by the theory of linear object class, getting the optimal coefficient of reconstruction by minimizing the error between the reconstructed face image and the primitive face image and reconstructs the face shape by this optimal coefficient. Then takes the reconstructed face shape as the initial model position of the active shape model (ASM) ,so get the initial parameter information, from this initial position of the model, we adjust the model parameter by learning to reduce the distance error between the model and the target profile until the model converge to the accurate position of the feature points. This method changes the strategy of model selection in the traditional ASM, by this method the model can converge to the target profile more exactly and rapidly.(2) A mixed feature extraction method base on optical model and temporal template is proposed. In this method, we firstly extract the optical feature vector of the feature points which were located in the above chapter. Then we extract the temporal feature of these feature points by sectional temporal template. Lastly, we fuse two feature vectors by the strategy of Canonical Correlation Analysis and get a set of Canonical feature. This method extracts and fuses two kind of features of the same face image sequence, it aggregates the merit of the two kinds of feature and eliminates the redundancy between them.(3) A facial expression recognition method based on feature fusion and HMM is implemented. The single feature has its limitation when denote the expression. Based on the multiple feature extraction and fusion we get a set of Canonical Correlation feature and set the fusion feature as the input of the HMM classifier then get the classification result.(4)A prototype system of facial expression recognition based on HMM is designed and implemented by object oriented technology. This system consist four modules that is face image preprocessing facial expression feature extraction and fusion, expression classification. It proves the validity of our method.