Research On Facial Expression Recognition

A full face expression recognition system can be divided into some parts: face feature detecting and locating, face segmentation and normalization, face expression feature abstraction. This paper mainly studies feature extraction and facial expression classification of static image of face.In this paper, we introduce background and review the history and the recent development of facial expression research. What's more, based on the actuality of domestic and international in recent years, we describe a survey of facial expression recognition's methods in terms of feature extraction and expression classification. Then introduce a face detection and positioning method, and have made some simulation experiment. This paper presents two improved methods on feature extraction: one method is 2DPCA which applied in enhanced feature images and another is F-2DPCA which applied in complex field. In order to improve the correct expression classification rate, first of all, introduce the weighted Mahalanobis distance classifier to classify and introduce the fuzzy C-means clustering to tectonic expression templates, then use improved particle swarm optimization(PSO) algorithm to optimize the initial value of fuzzy C-means clustering .The main conclusions are as follows:(1) The improved method of feature extraction proposed in this paper is 2DPCA which applied in enhanced feature images. This method effectively reduces the interference in same parts of the different facial expression and enhances the contrast of feature, experiments prove that the method can effectively improve correct expression recognition rate.(2) In this paper, F-2DPCA which is applied in complex field is proposed for more comprehensive extraction of useful image information, and also considers simultaneously differences in one class and in different classes, experiments prove that the method improves correct expression recognition rate effectively.(3) At the basic distance classifier research, the weighted Mahalanobis distance is applied to the classification of facial expression, the method taking into account the discrete degree of baseline template and the projector distribution of samples feature vector on the main vetor is been more precisely described.(4) Improved the inertia weight of basic particle swarm optimization, the method make the algorithm in the early stages in high convergence rate, also in the latter part the local search ability of algorithm is good . Not only retains advantages with increasing inertia weight and decreasing inertia weight particle swarm optimization, but also overcomes disadvantage, and achieves relatively good experimental effect.(5) Fuzzy C-means clustering is applied in facial expression recognition for facial expression templates construction. Because fuzzy C-means clustering relys on initial value, use the improved particle swarm optimization in this article fuzzy to optimize the initial value of fuzzy C-means clustering. This way overcomes the the shortcoming that fuzzy c-means clustering algorithm rely on the starting value excessively and easy to fall into the partial minimum.