Research And Application Of Face Recognition System Based On Convolutional Neural Network

Face recognition technology is a biometric technology that verifies face feature information as a criterion.Face recognition has been widely used in financial services,criminal investigation and other fields,and plays a vital role.In practical applications such as criminal investigation,public security face recognition or electronic payment face recognition,the collection of face images is carried out in the outdoor environment,often accompanied by more complex different lighting situations under natural conditions.On the existing face recognition technology,this thesis focuses on the complex illumination problems affecting face recognition.First,a light recovery method is designed to turn complex light to slight and moderate light changes and reduce the light impact of face images.Secondly,the needle deep face is specially affected by light changes,using the low-frequency discrete cosine transform(DCT)to normalize the images under different light conditions,to improve the discrimination of deep face features.First,LBP operators with very robust in illumination are introduced in CNN networks,then expanding the data improves the network generalization performance.This thesis focuses on the design of a face recognition algorithm(Local Binary Patterns and Data Expansion CNN,LECNN)that combines local binary mode LBP and data augmentation techniques into CNN networks.The input terminal uses the LBP feature face with stronger anti-interference ability on light characteristics instead of the face original image,and expands the LBP characteristic face data set by confrontation network generation in training conditions to effectively improve the CNN face recognition performance.Secondly,the image is transformed by low frequency discrete cosine transform(DCT).According to the size of correlation,the coefficients of DCT are calculated adaptively,and the DCT coefficients are normalized by using nonlinear modifier.Thus,the illumination in face image can be normalized adaptively.According to the change of illumination in the image,the number of DCT coefficients modified by nonlinear modifier is used.The method can adaptively standardize the illumination from facial images,thus enhancing the discrimination of depth features.The method achieves 100% correctness on all sub sets of YALE B data sets,CMU PIE,and sub sets 3 of extended YALE B data sets.Good results were also achieved in the rest of the face data sets.Finally,an intelligent face recognition system is designed and implemented using the face recognition algorithm proposed in this thesis.By using the Flask framework,the network model trained by this algorithm is encapsulated into an interface,which is applied to the intelligent face recognition system designed independently in this thesis,and tested in practical application.