Research And Implementation Of Feature Extraction And Classification Method For Facial Expression Recognition

Expression is the basic way for humans to express emotions and is an important means of nonverbal communication.In daily communication,people can identify the other person's psychology and conduct more effective communication through accurate and detailed analysis of facial expressions.With the continuous development of computer science and technology,human-computer interaction technology has increasingly become a research hotspot in the field of artificial intelligence.As an important part of human-computer interaction technology,facial expression recognition has received more and more attention because of its wide application prospects.The recognition accuracy of the current facial expression recognition system is not high,and the complex face image detected in the real scene cannot be effectively processed.This paper will study and analyze the feature extraction and classification and recognition of facial expressions,and design a new facial expression recognition model to achieve high precision and high robustness of facial expression recognition.Since deep learning methods have been widely used in pattern recognition and classification tasks in recent years,the convolutional neural network method has shown strong performance in the fields of image classification and face recognition,and is suitable for the expression extraction and expression classification links to be studied in this paper.Therefore,the facial expression recognition model designed in this paper will improve the facial expression recognition effect based on the convolutional neural network.The specific work and main contributions of this paper are as follows:(1)Additional redundancy features that are generated when facial expressions are acquired due to different postures,angles,illuminations,and occlusions.In order to avoid the use of convolutional neural networks for facial expression recognition,these facial features are extracted.This paper proposes a method based on multi-task learning to extract facial expression features.Using the facial landmark location task to implicitly promote the facial expression recognition feature extraction task,the multi-task learning framework is used to input the complete face sample image,and the facial expression recognition task and the facial landmark location task are integrated into the same convolution.In the neural network training model,after the feature extraction is performed,a preliminary expression classification recognition is performed on the acquired global advanced features and the facial landmark positions in the sample image are obtained.(2)Considering the local features generated by different feature parts of the face is an important basis for judging complex expressions,this paper will adopt a multi-network cascading framework.The multi-task learning framework combining facial expression recognition and facial landmark location is used as the first-level network for preliminary expression classification and feature landmark localization.In the second-level network,according to the facial landmark location result of the first-level network,input each The feature points are located near the image area for local expression classification.Finally,the expressions are accurately identified and classified by fusing the output of the multi-level network.(3)Based on the above research,this paper designs a set of facial expression recognition system.The system carried out multiple sets of comparative experiments on the FER2013 facial expression database and JAFFE library.The experimental results verify the effectiveness of the proposed method.