Research On Face Recognition Based On Feature Fusion

In recent years,with the advent of the information age,humans are paying more and more attention to information security.Facial recognition technology is widely used in transportation,public security,national defense and other fields.But the accuracy of facial recognition needs to be further improved.The paper conducts in-depth research on the problem of low accuracy of face detection and recognition and proposes an image decomposition method based on the optimal proportion of singular values,a face detection and recognition method based on feature fusion,and a dynamic face recognition method based on living body detection are proposed.The main research contents of this article are as follows.(1)Aiming at the problem of high dimensionality of face image data,an image decomposition method based on the optimal proportion of singular values is proposed.Firstly,the singular value decomposition method is used to reduce the dimensionality of the original face Image.Secondly,the face images with different proportions of singular values are used as input,and face detection experiments are performed using face detection method based on the Haar features and HOG features.Finally,the optimal proportion singular values is determined according to the difference between the detection rates of two face detection methods under different proportion of singular values.The experimental results show that:on the premise of obtaining effective face feature information,the optimal proportion of singular values determined by the method proposed in the paper on the ORL face database is 98%,and the spatial complexity of face image is reduced by 78.5%.(2)To solve the problem that the rate and accuracy of traditional face detection methods need to be improved,a face detection method based on feature fusion is proposed.The method uses the idea of integrated classification,which fuses the face detection methods of Haar features and HOG features to establish an improved face detection model.Comparative analysis of the experiments on the ORL database can be concluded that the accuracy of the paper method is improved by 10%,2%and 7%compared with the face detection methods of Haar,HOG and LBP features;and in terms of time efficiency,compared with the face detection methods of HOG and CNN have been improved by 14.6%and 99.2%respectively;The experimental results on the LFW database show that the accuracy of the paper method is improved by 15%,0.5%and 12%compared to the face detection methods of Haar,HOG and LBP features,and in terms of time efficiency,the paper method is improved by 41%and 99.4%compared to the face detection methods of HOG and CNN.(3)In order to solve the problem of low accuracy of the classification decision algorithm for face recognition,a face recognition method based on feature fusion is proposed.Firstly,based on the optimal proportion of singular values decomposition method and MTCNN's face alignment method,the face images are aligned.Then,in order to improve and perfect the face recognition model based on KNN and SVM,the MTCNN-KNN face recognition model based on the optimal K value and the MTCNN-SVM face recognition model based on the optimal kernel function are constructed.The experimental results show that the MTCNN-KNN face recognition model with the optimal K value increases its recognition rate by 10%,and the MTCNN-SVM face recognition model with the optimal kernel function increases its recognition rate by 2%.(4)To meet the needs of a wider range of applications,there is a pseudo-identity attack method in the field of face recognition.The paper studies in the field of dynamic face recognition and proposes a dynamic face recognition method based on live detection.Based on the in-depth study of blink detection,this method designs a live detection method for open mouth detection,which combines open mouth detection and blink detection to achieve dynamic face recognition.The experimental results show that the dynamic face recognition method based on living body detection provides real-time and reliable security for identity authentication and has engineering feasibility.