Research On Deep Super-Resolution Reconstruction And Recognition System Of Face Images

Face recognition is an important technology of identity verification and recognition using human facial features,which has a wide application prospect in daily life.During recent years,many face recognition algorithms have been proposed.These algorithms have achieved good recognition results on constrained face image data sets.However,in real application scenarios,due to the relatively backward configuration of video acquisition equipment,bad natural environment,serious noise interference and other external factors,the accuracy of existing face recognition algorithms in these complex conditions has declined.Therefore,it is very important to improve the accuracy of face recognition under the condition of low resolution,and it also has great application value in real life.This paper focuses on the super-resolution reconstruction of low resolution face image and the key factors of face recognition algorithm.Combined with the traditional spectral reconstruction algorithm and deep learning,this paper proposes a face deep super-resolution feature reconstruction and recognition algorithm,and applies the algorithm proposed in this paper to design and implement a face recognition system to verify the effectiveness of the algorithm proposed in this paper.The main research work of this paper is as follows:(1)A face image super-resolution reconstruction algorithm based on fractional face representation is proposed.Noise and other interference information will have adverse effects on face image super-resolution reconstruction.The fractional representation of face image can effectively suppress these interference factors and enhance the robustness of face image to noise,light intensity and expression changes.In this paper,the fractional representation of face image is used as the input of convolution neural network.The residual feature image is extracted and combined with the low resolution face image to get the reconstructed super-resolution face image.In the experiment,we first determine the optimal fractional order parameters through quantitative experiments,and then prove the effectiveness of the algorithm on Celeba and AT&T face image databases.(2)A face recognition algorithm based on multi feature fusion is proposed.The robust recognition of low resolution face images usually requires a variety of face image features.Based on this,this paper proposes a feature fusion method,that is,the super-resolution face image constructed in this paper is fused with the face image features obtained under different fractional parameter constraints,and the fused face features are obtained and used for recognition.Different face recognition loss functions are used in the experiment,and then the effectiveness of the algorithm is proved on Celeba and AT&T face image databases.(3)Using the proposed face image super-resolution reconstruction and multi feature fusion algorithm,a face depth super-resolution feature reconstruction and recognition system is designed and implemented.The system mainly includes image acquisition,face detection,face image preprocessing,face image preprocessing,face image super-resolution reconstruction and face recognition modules.After testing each functional module of the system,it is proved that the system can effectively complete the task of face image super-resolution reconstruction and recognition.