Research On Facial Expression Recognition Based On Multi-feature Fusion

The study of facial expression recognition has a long history,and there generated many good algorithms of features extraction and classification.However,the recognition rate is still restrict the practical application of expression recognition.To solve the still existing problem of facial expression recognition,we have researched the feature extraction and expression classification recognition in facial expression recognition.The main contents and innovations are as follows:(1)We proposed a facial expression feature extraction method based multi-feature weight fusion(MFWF).The single feature information can not describe the image accurately,which affects the recognition rate of the image.In order to accurately express the image information with the characteristic information and improve the recognition rate of the image,the concept of feature fusion is introduced for the extraction of facial expression features.Firstly,to determine the local interest area in face;Secondly,using a variety of feature extraction methods obta'ins multiple feature information of face;Finally,the different features are weighted due to different performance of facial expression in different facial regions.With the method of serial fusion,the extracted feature is connected into one line as the final characteristic information of the image.In order to verify the effectiveness of the MFWF method,the JAFFE database is used as the data source,and the classification recognition result of fusion features and non-fusion features are compared.The experimental results show that the feature fusion can improve the representation ability of the image and improve the recognition rate of the expression.Especially in the three classifiers of LDA,DT and DDA,the effect is the most obvious.(2)We proposed a facial expression recognition method based on quadratic optimization choice(QOC)ensemble classification model.Aiming at the defects of the single classifier,the idea of ensemble learning is introduced to generate an ensemble classifier model for facial expression recognition.Firstly,for a large number of base classifiers,they are ranked from large to small according to their classification,and to determine the base classifier which to ensemble in accordance with the corresponding stop criteria;Secondly,for the classifier ensemble to be determined,the ensemble model cluster is generated according to the ensemble rules;Finally,the performance of the ensemble model cluster is tested,and the integration model is sorted from large to small.the first ensemblemodel has been ranked as the final ensembleclassifier model.In order to verify the performance of the QOC method,we use the JAFFE database as the test data set,and use the ensemble rules of the maximum,minimum and mean as the contrast model.The experimental results show that compared with the base classifier,the classification recognition performance of the ensemble classifier is significantly improved,and the SRM ensemble model increases the average recognition rate of sad expression from 78.89%to 96%;Compared with the non-selection ensemble model,the LFER ensemble rule in QOC method is better than other ensemble rules;Compared with the selection ensemble model,the selection strategy in the QOC method can effectively improve the recognition rate of the ensemble model.