Research On Facial Expression Recognition Algorithm Based On Local Binary Pattern

The facial expression is as the carrier of conveying the emotional information,which has played one indispensable role in the course of man-machine interaction.The traditional expression characteristics extraction method has poor flexibility,and the extracted characteristics are easy to be affected by the complex environment.It will require a large number of training data if using the deep learning method to extract the characteristics,the insufficient data volume will lead to the model to generate the overfitting,and training the deep neural network is time-consuming.Therefore,the effective methods of researching the expression characteristics to obtain and enhance the expression recognition rate are the key points for the facial expression recognition technology.Because LBP features are susceptible to noise interference,and LBP features are difficult to describe gray changes by comparing gray values of individual pixel points,the overall structure of expression images is ignored,resulting in the loss of some feature information.An improved LBP feature extraction algorithm by extending the neighborhood size of LBP operator is proposed in this paper.The algorithm uses the weighted gray value of the local neighborhood pixels to replace the gray value of a single neighborhood pixel.The weighted gray value of the neighborhood pixels and the gray value of the central pixel are calculated to obtain their absolute values,and then the average value is used for LBP coding to extract expression features,thus strengthening the connection between each pixel in the expression image.According to the classification of facial expression features,considering that SVM has strong nonlinear classification ability,an improved LBP feature is combined with SVM classifier to design a facial expression recognition method.F irstly,the improved LBP algorithm is used to extract feature vectors from preprocessed expression images,and then SVM classifier is constructed by one-to-one strategy to learn the obtained expression features.The open expression databases JAFFE and CK+are used to cross-verify the method.The simulation results show that the recognition rate of the method is 90.8%and 96.2%respectively on the two data sets,On JAFFE databases,the recognition rate with the existing algorithm is improved by 1.3%.Targeting at the problems of small number of available expression samples and insufficient feature information extracted by deep neural network,an improved face expression recognition method combining LBP feature and CNN model is designed.In this method,the improved LBP facial expression feature matrix is directly input into CNN model for training,which enhances the facial expression feature information extracted by CNN.JAFFE and CK+expression databases were used to cross-verify the method.Simulation results show that the recognition rate of the method is 93.5%and 96.4%respectively.On CK+databases,the recognition rate with the existing algorithm is improved by 4%.