Human Facial Expression Recognition Using Angles Geometric Features

With the rise of artificial intelligence,facial expression recognition has received more and more attention as a part of emotional expression and has become a research hotspot.In human sentiment analysis,expression is a very intuitive feature that expresses people's psychological state.In the process of human-computer interaction,the judgment on these characteristics can accurately provide valuable data and information.In addition,as a challenging research topic in the field of pattern recognition,facial recognition uses traditional image processing algorithms in the process of feature extraction and recognition,and neural networks and deep learning methods have also begun to be used in large quantities.Some researchers in the field have done relevant research.How to extract more effective facial features and improve the recognition rate is still an important work in current facial expression recognition.This paper first establishes an expression database,then proposes a feature-based geometric feature method on feature extraction.Finally,it classifies and recognizes by Support Vector Machine(SVM).Specific research work includes:(1)Establish an expression database to increase the number of data samples.Currently,most existing studies used existing databases such as CK+,JAFFE,etc.These databases have been effectively verified.In this paper,the latest 3D camera was used to record the expression database,and corresponding experiments were performed.(2)Analysis of commonly used feature extraction methods,due to the appearance characteristics of each person,the magnitude and manner of expression change is not the same,the general feature extraction is to first cut the image,normalized and other pre-processing.Reduce the impact on extraction and recognition.Based on the feature points,this article focuses on the analysis of the trends and characteristics of expression changes and focuses on analyzing the key changes of different expressions and converts the changes in of feature vectors that can be calculated as the basis of expression recognition.(3)The number of data frames of the image sequence is unified because the length of data required in the support vector machine is the same.The method adopted is to complete the number of frames of all image sequences of a fixed value.The specific method is to delete the frames of the first frame of the data longer than the fixed value,and to shorten the frame from the first frame of the data.Increase,the increase in value is the average before and after,this is to reduce the impact on the part of the expression change.Then,the data is normalized to reduce the difference in the appearance of the same eigenvector on different expressions because of different appearances,changes from into amplitude,and methods,which affects the feature extraction and recognition.After the above steps,the dimension of the feature vector is extended.Finally,using the support vectors vectored machine to identify,and study the impact on different kernel functions and recognition methods on the recognition rate.(4)There are up to 60 feature points in the selected facial expression database,and correspondingly,a considerable number of feature vectors can be combined.The starting point of this article is to start with the most intuitive expression judgment and construct as many feature vectors as possible.The selection of feature vectors is performed on the actual features of expression changes represented by these feature vectors so that the recognition effect is best.On the basis of the feature points,various feature vectors are combined and then identified.This avoids the need for step-by-step feature extraction through image processing to a certain extent and is more adaptable to different recognition environments.