The Research And Implementation Of Feature Extraction Of Facial Expression Recognition

Facial expression recognition is an emerging research and have broad application prospects in many other fields. Facial expression recognition system is divided into three stages: preprocessing, feature extraction and facial expression classification. In this paper, the research purposes is the expression recognition by facial expression, we analysis and optimization facial expression feature extraction algorithm, and designed to improve the efficiency of facial expression recognition.Feature extraction method of facial expression recognition is divided into feature integration and feature reduction. Feature fusion that is characterized by significant image feature extraction in accordance with specific rules. Feature dimension reduction that is characteristic data backbone information selection, they directly affect the expression recognition accuracy. In the expression feature extraction stage, due to changes in facial expression contains a wealth of texture information, and different facial expressions to show the changes in different scales, so usually use the Gabor filter expression to do feature extraction.this paper presents a feature fusion method to solve multi-directional multi-scale image features problem of high dimensionality of Gabor wavelet extraction. we do the fusion rules integration according to the specific characteristics feature images of the same scale in different directions and guarantee eliminate redundancy under the premise, to a certain extent, reduce redundancy, improve accuracy characteristics. In the Dimensionality reduction stage of expression feature, this paper presents an analytical algorithm(LPPCA),which is based on local principal components. And this algorithm combines the advantages of principal component analysis(PCA) and partial projection maintain superiority(LPP) algorithm,that is maintaining both locally geometry information data information and also the variance information of the data. That is to say, It has both global image feature extraction on local features of the image also has some ability to extract.and can identify different expression characteristics in the case where the feature dimensions decrease, further improve the effectiveness of facial expression feature extraction.In this paper, we research on the Japanese female facial expression database(JAFFE) and classify the experiment by support vector machine(SVM). The results of this experiment show that the improved algorithm, feature fusion algorithm can improve accuracy and increase the efficiency of facial expression recognition to identify in a certain extent; And the characteristic dimension reduction improved algorithm(LPPCA) not only to obtain information on global features but also obtain local features of being able to obtain more comprehensive information on features. Thus, doing the basic features of the improved algorithm converged on feature dimension reduction, can further improve the expression recognition accuracy to 96.86%. But the results also show that there is slightly operational loss in efficiency on improved characteristic dimension reduction algorithm. Because of its higher computational time complexity. However, for demanding real-time facial expression recognition but expect high accuracy applications, LPPCA algorithms has its advantages to recovery more complete feature information.