Research On Facial Expression Recognition Based On Improved HOG Features

Facial expression recognition is one of the widely-used human-computer interaction methods,and its related theories and technologies need further study.Although the technology related to facial expression recognition has been greatly developed,the robustness and accuracy of its algorithm need to be further improved.In the face recognition algorithm based on image operators,the extracted texture features are relatively rich,but they are prone to overfitting.The geometric features extracted through feature positioning points can just make up for this shortcoming.Although the related facial expression recognition algorithm based on deep learning has achieved good results,it will also appear overfitting during the training of specific models.Aiming at these problems,this thesis studies the facial expression feature extraction,the main research work is as follows:1.An improved HOG feature extraction algorithm is proposed.To address the shortcomings of traditional HOG feature extraction process.In this thesis,the grayscale information of diagonal pixels is added to HOG feature extraction,which makes the extracted grayscale edge information more abundant.In addition,the optimized HOG characteristics under different parameters were tested.2.A facial expression recognition algorithm based on texture features and geometric features is proposed.Although the expression features extracted by traditional image operators are relatively rich,they often appear to be over-fitted because they contain more redundant information.For this problem,firstly,the facial features area is located,and the texture features of the localized area are extracted.Secondly,the ERT feature point location algorithm is used to accurately locate the facial expression feature points,and useful geometric expression information is extracted according to the located facial expression feature points.Finally,the extracted geometric expression information is merged with the optimized facial expression HOG texture features,and an expression recognition algorithm based on geometric features and texture features is obtained.3.The deep learning algorithm used for facial expression recognition is optimized.Deep learning will also produce a lot of overfitting results in the process of facial expression recognition.To solve this problem,this thesis firstly extracts the edge information of the input image,and secondly studies the convolutional neural network model of related literature,and optimizes the convolutional network model by combining the methods of partial literature extraction to reduce the occurrence of overfitting.Experimental results show that the optimized deep learning model has better robustness.