Research On Facial Expression Recognition Algorithm Based On LBP And HOG Feature Fusion

Expression is an important way to convey feelings in human daily life,which can intuitively reflect the psychological state of human beings.With the continuous development of computer technology and artificial intelligence,expression recognition has become a hot topic at home and abroad.Through reading a large number of references and books in the early stage,this thesis deeply studies the feature extraction algorithm and the algorithm of different feature fusion in the expression recognition process.The original expression information can not be completely described for a single feature.This paper proposes a feature fusion algorithm for weighted fusion of different regional features.The main research contents and work of this paper are as follows:(1)In this paper,the Adaboost face detection algorithm is used to screen out the face region from the image to be recognized,and the position of the human eye is located by horizontal and vertical projection.According to "three foreheads and five eyes" The face model cuts out a pure emoticon image.After normalization by scale and normalization of gradation,an expression image with less interference information is obtained.(2)For the traditional local binary mode(LBP)algorithm,only the gray value of the central pixel in a computing unit is subtracted from the gray value of the neighboring pixel,but it ignores the relationship between the neighboring pixels.In this paper,an improved LBP feature extraction algorithm is proposed.The improved LBP algorithm introduces a threshold value M,which is the absolute value of the gray value of the central pixel and the neighboring pixel,and then each the absolute values are added together and averaged.Finally,the central pixel and the domain pixel are compared to an absolute value and compared with a threshold M,which is greater than the threshold coded to be 1 otherwise 0.The effectiveness of the improved algorithm is verified by experiments on multiple expression datasets.(3)For the problem that the single feature cannot accurately describe the original featureinformation,this paper uses a different region weighted feature fusion algorithm.The LBP algorithm is used to extract the texture features of the face region.The HOG algorithm is used to extract the edge and shape information of the eyebrow region and the mouth region.and the three regions are given different weights and then merged.The fusion feature information is sent to the classifier for classification and identification.The effectiveness of the feature fusion strategy is verified by experiments on different data sets.(4)In this paper,support vector machine(SVM)is used to classify and recognize expressions.The processed fusion features are sent to the SVM classifier for classification and identification,and the K-nearest neighbor classification algorithm(KNN)is compared on different data sets to verify that the SVM algorithm is better than the KNN algorithm.