Research On Robust Facial Expression Recognition Method And The Implementation Of System Under Partial Occlusion

Facial expression contains a wealth of personal emotional information, recognizing the facial expression automatically has broad application prospects in human-computer interaction, intelligent security and psychological analysis and so on. At present, most research on facial expression recognition is focus on images without occlusion. However, facial occlusion usually occurs in real life, which will have a negative effect on recognition rate and robustness of the algorithm. Therefore, the more robust algorithm for facial expression recognition under partial occlusion has become a research focus in intelligent human-computer interaction, affective computing, pattern recognition and other fields.This paper studies on the robust expression recognition method under partial occlusion. Firstly, existing methods are used to complete face detection and image pre-processing. Then, in order to improve the classification accuracy and robustness of facial expression recognition algorthms under partial occlusion, this thesis is mainly to have algorithm research in feature extraction and expression classification. The main work completed in this thesis is listed as follows:(1)A novel expression feature extraction algorithm based on Gabor wavelet and GLCM is proposed. At first, we design a method to extract the block Gabor feature statistics to overcome the shortcomings of high dimension Gabor features, this method can not only remain the spatial characteristics of face organ, but also reduce the dimension of Gabor features. Then, GLCM is firstly introduced into expression recognition field to make up for the deficiency of block Gabor feature, in which the association between pixels is absent. The experimental results show the high robustness, low feature dimensions, short classification time and high recognition accuracy of the proposed novel approach under different types of occlusion.(2)Design and implement an optimal multi-class support vector machine(SVM) expression classification algorithm. Based on multi-category support vector machines theory, the algorithm uses the expression features extracted by step(1) as the input, and selects radial based function(RBF) as the kernel function, then determines the optimal kernel parameters g and the penalty parameter C by grid search method, leading to an optimal SVM model. Finally, we use this model to complete the expression recognition of the test image. The experimental results show that, compare with the baseline classification algorithm, this algorithm can overcome the negative effects caused by individual differences of expression in different data environment, and achieve higher recognition accuracy.(3)A facial expression recognition system under partial occlusion is designed and implemented. Based on C/S framework, The system transforms the theoretical achievements mentioned above into C language modules, contributing to an integrated robust facial expression recognition system. System performance test results show that, the system can identify the occlusion image which from the professional expression library with a good recognition effect, the recognition rate up to 80%. As for the occlusion image collected in practical complex environment, the system shows a rough classification capability, which indicates that the system has certain practicability. Design and completion of this system can provide some technical reserve for facial expression recognition from experiment towards practical application.