The Research And Implementation Of Facial Expression Recognition Based On Static Images

In today's society,where the computer technology has been rapidly developed,facial expression recognition plays an increasingly important role in various fields,and gradually develops into a very hot topic in the field of scientific research.In this paper,the study of the facial expression recognition algorithm based on static image is from two aspects: global feature extraction,such as Gabor wavelet transform,LBP operator,LPQ operator,and local feature extraction,such as SIFT operator.The main work is as follows:(1)The application of the face expression recognition algorithm that fuses LBP and LPQ operators based on Gabor transform is studied.LBP operator and LPQ operator do well in extracting local features of expression images,but have the shortcomings that can only be described on a single scale.The Gabor wavelet has a unique advantage in directional selection,scale selection and robustness to light changes.In this paper,the advantages of both are combined together.Firstly,the Gabor feature is extracted from the face image,and then the LBP operator and LPQ operator are coded for each Gabor convolution image,and the LGBP and LGPQ features which are more capable of characterizing facial expression are obtained and then fused.In order to further reduce the dimension of the feature and improve the computational efficiency,the PCA+LDA method is used for the fused feature.The experimental results show that the method based on Gabor transform can effectively improve the accuracy of expression recognition,with an increase of more than 20%.(2)The application of the improved bag of words(BOW)model for face expression recognition is studied.Due to the lack of enough distinction in the visual dictionary constructed for the whole expression image,a new method of visual dictionary construction is adopted in this paper.The idea of the region of interest(ROI)is introduced,then the DSIFT feature is extracted and the BOW model is constructed for the eye and mouth regions only,because they both have a higher contribution for facial expression recognition.Then the two vocabulary frequency histograms are linearly combined for face expression recognition.The results show that the accuracy of the improved method is about 3% higher than the traditional.(3)The application of the improved spatial pyramid matching(SPM)algorithm for face expression recognition is studied.By adding the spatial structure information of the feature,SPM can effectively improve the performance of the BOW.In this paper,the BOW and SPM are applied to the two ROI respectively,and the obtained features are merged for easier discrimination.Due to the large dimensionality of SPM feature,the histogram intersection kernel(HIK)and the method based on Relief F are introduced.The HIK method can improve the accuracy by about 1% while reducing the dimension.The method based on Relief F can maintain the accuracy while reducing the classification time by 20%,which improves the efficiency of classification and is of great significance to the large-scale processing of facial expression images.In this paper,JAFFE and Cohn-Kanade+ are applied on the algorithms mentioned before to verify the generalization and effectiveness of them.The experimental results show that the improved algorithms have a better classification effect on both two expression databases,indicating that the generalization of them is better.