Facial Expression Recognition Based On Gabor Features And Handwritten Chinese Character Recognition

In order to facilitate a more intelligent, natural and friendly mode for communications between computers and human, novel human computer interface has become a very active research area in computer vision and pattern recognition. There are many approaches have been proposed for facial expression analysis, Chinese character recognition and so on. This thesis studies the problems of Gabor feature extraction, feature reduction and classifier, facial expression recognition system, and handwritten Chinese character recognition.Firstly, we studied the properties of the Gabor filter. The 2D Gabor filter can be regarded as a product of an elliptical Gaussian and a complex plane wave, and then we analyzed the properties of the Gaussian function and the complex plane wave in details. The Gabor feature representation and Gabor filter bank are also introduced.Secondly, there are two classical approaches to reduce the excessive dimensionality by combining features. One approach, known as Principal Component Analysis or PCA, seeks a projection that best represents the data in a least-squares sense. Another approach, knows as Fisher Linear Discriminant or LDA, seeks a projection that best separates the data in a lease-squares sense. Based on the methods, we designed the framework of the classifier, and paid more attention to the details of distance classifiers with different distance metric.Thirdly, we focused on the Facial Expression Recognition (FER) system, including the preprocessing procedure, feature extraction and classifier. This paper proposes a new local Gabor filter bank to overcome the disadvantage of the traditional Gabor filter bank, which needs a lot of time to extract Gabor feature vectors and the high-dimensional Gabor feature vectors are very redundant. In order to evaluate the performance of local Gabor filter bank, a FER system based on Gabor feature is presented. The FER system extracts Gabor feature of pure facial expression images, then it uses a two-stage method PCA plus LDA to select and compress the sub-sampled Gabor feature. Distance classifiers are used to recognize facial expression. Experimental results show that the method is effective for both dimension reduction and recognition performance. The novelty of the method is to select partial Gabor filter bank with part of m scales and n orientations to extract Gabor feature. For JAFFE database, the best average recognition rate of 97.33% was achieved, which indicated this method was suit for facial expression analysis.Finally, we compared multiple directional features with Gabor features in terms of the recognition performance for handwritten chinese character recognition. Based on the comparison, Gabor feature extraction with multiple scales was presented, which can achieve higher recognition rates.