Research On Facial Expression Recognition Algorithm Based On HOG Difference Weights

This article first introduces the significance of expression recognition research and the current situation at home and abroad.Then pre-process the emoji image.Secondly,for the problem of HOG(Histogram of Oriented Gradient)algorithm that only extracts a single angle gradient and ignores the diagonal gradient,a multi-angle HOG(Ma-HOG)feature extraction algorithm is proposed.In order to obtain the local details and global information of the image,a multi-scale Ma-HOG(Mas-HOG)is proposed.Aiming at the contribution of facial regions to facial expression recognition,a Mas-HOG algorithm based on differential weights is proposed.Finally,comparative experiments are performed on different classification algorithms to select the most suitable classifier for this algorithm.The main research contents of this article are as follows:(1)Obtain ENM(Eye Nose Mouth)regions of interest and reduce the feature dimension.According to the main expression characteristics of facial expressions,the face is divided,and the eyes,nose,and mouth of the face are obtained.By constructing the region of interest,the amount of input data can be reduced and the feature dimension can be reduced.(2)A Ma-HOG expression feature extraction method is proposed to obtain multi-angle gradient information.For the HOG algorithm,only the horizontal and vertical gradient calculations are considered.In this paper,the diagonal gradient information of the image is calculated,which together with the original HOG algorithm gradient information constitutes the Ma-HOG feature of an image.(3)A Mas-HOG feature extraction method is proposed to extract multi-scale Ma-HOG features.In order to solve the problem that the information of the facial expression area of the human face is not comprehensive at a single scale,the multi-scale lower-level expression information is obtained by constructing a multi-scale spatial pyramid.Combined with Ma-HOG features,Mas-HOG facial expression features are formed.The results of multiple experiments show the rationality of the algorithm proposed in this paper and effectively improve the expression recognition rate.(4)A method of calculating differential weights is proposed to weight each region of the face and extract the Mas-HOG features of differential weights.In view of the different contributions of facial regions to expressions,the difference weight of the ENM region is obtained by acquiring the difference image of the original image and counting the changes in the difference image pixels.Combined with the Mas-HOG algorithm to form the differential weight Mas-HOG algorithm.Through multiple sets of comparative experiments,a higher recognition accuracy than other HOG improved algorithms is achieved,thus verifying the superiority of the proposed algorithm.(5)Based on experimental comparison under different classifiers.Introduced common classification algorithms,and compared the algorithms in this paper with KNN(K-Nearest Neighbor,K nearest neighbor classification),BP(Error Back Propagation Training)neural network,and SVM(Support Vector Machine)classifiers.experiment.The experimental results show that the feature extraction method proposed in this paper is more suitable for classification and recognition under SVM.Figure 37 Table 12 References 59...