Research On The Algorithm Of Facial Expression Recognition Based On Dynamic Fuzzy Neural Network

Facial expression is an important interactive tool for expressing emotion information, which has potential application value in high demanding intelligent human-computer interaction. As an important part of the human-computer interaction technology, facial expression recognition involves several research fields, such as computer vision, emotion computing, image processing, intelligent control and pattern recognition. Recently, it becomes the research hot and difficult at home and abroad. At present, automated facial expression recognition system is mainly to solve three key problems:face detection, facial expression feature extraction and classification. With the development of nearly 40 years, researchers have proposed many facial expression feature extraction algorithms and classification for the facial expression recognition and made great progress. However, it still has a considerable distance for the demand of practical application and the problem solving completely, which makes us do further research on the issue of facial expression recognition.In view of the problem that the right recognition rate and learning efficiency of the existing expression recognition algorithms are relatively low, especially for the unobvious expression, we propose a new facial expression recognition algorithm based on Dynamic Fuzzy Neural Network. The algorithm mainly includes facial expression images preprocessing, facial expression feature extraction and dimensionality reduction, face classification and recognition. Firstly, we do the geometric preprocessing and gray preprocessing for the original facial expression images, which is ready for the expression feature extraction and recognition.Then, in the expression feature extraction, this paper proposes a kind of feature extraction strategy, which combines Gabor wavelet and two-dimensional Principal Component Analysis (2DPCA)+fuzzy linear discriminant analysis (Fuzzy LDA). Firstly, this algorithm extracts the Gabor features of the preprocessed image by the Gabor wavelet transformation, and then reduces the dimensionality of Gabor high-dimensional feature vector by the sampling method and 2DPCA. Finally, introduces fuzzy set theory to the traditional LDA method, extract the low-dimensional and most discriminant facial expression features by the Fuzzy LDA algorithm.Moreover, in the facial expression recognition, a classifier based on dynamic fuzzy neural network is constructed, which is used to classify and identify the facial expression input samples. In the process of recognition, we consider the extracted facial expression features of training set and the corresponding classes as the input and output data of Dynamic Fuzzy Neural Network classifier, which is used for design and train the DFNN. Then, we use the trained DFNN classifier to classify for the testing samples.Finally, this paper conducts simulation experiments to implement the above facial expression recognition algorithms in Japanese Female Expression Database (JAFFE) by MATLAB R2011a. And then, we make the contrast experiments with other facial feature extraction algorithm (Gabor+ PCA, Gabor+PCA+LDA, Gabor+2DPCA) and classification recognition algorithm (BP, RBF, LVQ, SVM, HMM). The experimental results show that the right recognition rate and performance of this algorithm proposed by this paper is better than others. Especially for the unobvious expression, its recognition rate is high, and the time required for identification can keep in the reasonable scope. So this algorithm obtains an ideal result.