Research On Facial Expression Recognition Algorithms Based On Bayesian Network

Facial expressions are the most direct way to convey emotions.With the development of science and technology,facial expression recognition is widely used in various fields.Bayesian Network(BN)has received much attention for its advantages in dealing with complexity and uncertainty issues.Aiming at the problem that the facial expression recognition process is more complicated and uncertain,a facial expression recognition algorithm based on Bayesian network is proposed.Based on the analysis and research of facial expression features,the feature extraction is performed by using the Histogram of Oriented Gradient(HOG)algorithm and the Constrained Local Model(CLM)algorithm.The support vector machine(SVM)is used to classify and obtain the action unit(AU)data set.For the problem of limited sample size when modeling the system,a BN parameter learning algorithm that convex optimization based on small data with expert knowledge(CSDE)is proposed.Apply the CSDE algrithm to facial expression recognition.The main contents of this paper are as follows:(1)The acquisition process of the facial action unit(AU)data set characterizing facial expressions was studied.Firstly,the influencing factors of facial expressions are analyzed.The CLM algorithm is used to locate the feature points,so as to obtain the geometric features of the facial expression images,which are used to characterize the shape changes of facial organs caused by facial muscle movements.Then,the HOG feature is extracted by the HOG algorithm and the dimensionality reduction is performed by using the PCA to represent the subtle changes of the local texture of the face;Finally,the two are characterized by feature fusion and normalization and classified by SVM to obtain the AU sample data set.(2)Aiming at the problem of BN structure modeling in facial expression recognition process,a face expression recognition BN structure model based on facial action unit(AU)is proposed.The structure determines the total number of nodes in the BN structure and the pointing order between the nodes according to the known AU sample data set and the detailed analysis of the facial expression and AU relationship,and the knowledge of the domain experts,and finally determines the facial expression.Identify the BN structure and provide a basis for facial expression recognition.(3)Aiming at the problem of low learning accuracy of BN parameter learning method under the condition of small data sets,a convex optimization parameter learning method based on small data sets and expert experience(CSDE)is proposed.When the BN structure is known,the qualitative expert experience is transformed into a set of constraints between the conditional probabilities of BN.Then the convex optimization method is introduced to estimate the parameters of BN model.The simulation results show that the algorithm not only does not depend too much on expert experience,but also makes full use of some inequality constraints of expert experience.To a certain extent,it makes up for the influence of insufficient data on the accuracy of parameter learning,and then completes BN modeling under the condition of small data sets.(4)Aiming at the low accuracy of facial expression recognition under limited dataset,a facial expression recognition algorithm based on CSDE parameter learning algorithm is proposed.The selected expression database is Cohn-kanade expression database.According to the feature extraction method mentioned in this paper,the expression features are extracted.After normalization and other processing,the action unit(AU)sample data set is obtained.Combining the relationship between facial expression and AU,the BN structure model of facial expression is constructed.The BN model is completed by parameter learning using CSDE algorithm.Finally,the joint tree reasoning algorithm is used to deduce the corresponding facial expression attributes.The experimental results show that the CSDE algorithm has a satisfactory effect in the application of facial expression recognition,which can realize facial expression recognition under the condition of small data sets.