Research On Algorithm Of Human Facial Expression Strength Estimation

In people-to-people communication,facial expressions,as a very important way of transmitting information,can convey information that many languages cannot convey.With the rapid development of artificial intelligence and the Internet of Things technology,if humans and robots can also communicate with each other through facial expressions and realize intelligent interaction between humans and machines,it will greatly promote the development of artificial intelligence technology.Therefore,the analysis and recognition of computer facial expressions has important research significance.In the facial expression recognition system,feature selection and extraction are very important links in the process of facial expression recognition.The more intrinsic the feature extraction,the higher the recognition performance.As the recognition accuracy has not yet reached the best,the existing feature extraction methods and the optimization of the classifier still have the potential for improvement.In addition,many researchers at home and abroad have made a lot of research on emotion recognition in recent years.From single-modal emotion recognition to multi-modal emotion recognition,most of these studies have distinguished six basic facial expressions.However,or the same facial expression,the intensity of the expression is different,and the emotional state displayed will also be different.Emotion recognition and analysis should not be limited to only a few coarsegrained classification results.Aiming at the above problems,this paper first proposes an expression recognition method based on Gabor filters.First detect and locate the face of the image,then use the Gabor transform that is sensitive to edge areas and robust to light changes to extract expression features,then use BDPCA + LDA to reduce dimensions,and finally use Euclidean distance and nearest The neighborhood method classifies the expressions and completes the expression recognition of the images.Then,an expression intensity estimation method based on fuzzy clustering is proposed.Based on a large class of emotion recognition,multiple frame image samples of "anger" emotions are taken from the end of the classifier,and 200 frames of test and inspection samples are selected from them.The selected continuous samples are fused and clustered to determine the initial cluster Number of class centers.Finally,the Adaptive Neural-Based Fuzzy Inference System(ANFIS)is used to score the identified angry expressions.The higher the score,the higher the degree of anger.In order to verify the performance of expression recognition method based on Gobor filter and expression intensity estimation method based on fuzzy clustering.Based on the JAFFE database and Ck + database,this paper conducts experiments on expression recognition and expression intensity estimation.The experimental results show that the Gabor + BDPCA + LDA expression recognition method designed in this paper can reach a maximum of 97.21%,and the expression intensity estimation method based on fuzzy clustering can reach a maximum of 83.34%.Experiments show that these two methods have good feasibility and robustness in real environment.