Research On Facial Expression Recognition And Data Enhancement

Since entering the era of artificial intelligence,facial expression recognition technology has been of great significance in the fields of human-computer interaction and computer vision.With the continuous integration and improvement of modern computer hardware and software,it has gradually become the focus of attention of researchers.This dissertation focuses on the research of feature-level data enhancement of static facial expression recognition algorithms and their databases.The specific research work is as follows:1.The research background and current status of facial expression recognition technology and data augmentation methods are expounded.Starting from the various steps of facial expression recognition,the key technologies and commonly used related algorithms of each link are described in detail.In addition,related theories about generating adversarial networks are also introduced,and some brief analysis and discussion of their derivative models are made.2.The shortcomings of the traditional feature extraction algorithm Local Binary Pattern(LBP)and Gradient Direction Histogram(HOG)in the feature description are briefly analyzed and improved.A fusion double coding local binary value is proposed.Facial expression recognition method of pattern(DCLBP)operator and absolute gradient histogram(HOAG)operator.This method first performs face detection on static pictures,and performs geometric transformation,scale normalization,and filtering on the detected face regions to overcome the adverse effects of illumination,pose,and noise on expression recognition;second,it uses DCLBP The operator extracts the local texture features of the face image,and uses the HOAG operator to extract the local shape features of the face image.Then,the two extracted features are fused using the typical correlation analysis method(CCA)to fully exploit the complementarity of the two features.And recognition performance;finally,support vector machine(SVM)is used for facial expression classification.The experimental results show that compared with the single feature recognition method and the cascade feature recognition method,the proposed method achieves better recognition results.The experiments on the Cohn-Kanade(CK)and JAFFE datasets have reached100% and 99.05%,respectively.The comparison of the recognition rate with other related methods also verifies its effectiveness.3.First,the problems of insufficient sample size,imbalanced categories and high costof constructing new databases in the existing facial expression database are studied.Aiming at the insufficient data enhancement of traditional generative adversarial networks,an improved star-shaped adversarial network is proposed(Star GAN)face expression picture generation method.This method first improves the structure of the model on the basis of Star GAN,and adjusts the loss function to solve the problems of blurring,blackening,etc.that occur in the local area of the image generated by Star GAN.Then,the mask vector is used to convert the The style is jointly trained with the style of another data set to achieve multi-style facial expression picture generation across data sets.The experimental results show that,compared with traditional data enhancement methods using image geometric transformation and pixel value transformation,the method in this paper not only increases the number of samples,solves the problem of data imbalance,but also enriches the features extracted by the model and increases the applicability of the model.