Research On Face Image Super-resolution And Recognition Algorithm Based On Deep Learning

Image super-resolution reconstruction technology refers to the technology o f reconstructing low-resolution images into high-resolution or super-resolution images.It is the same as image enhancement and is an image processing technology.Face image super-resolution reconstruction is an important branch of image super-resolution reconstruction.Low-resolution face images often lack facial texture feature information and contain more noise.How to reconstruct these missing texture feature information and effectively suppress noise has become an important problem in the task of face image super-resolution reconstruction.From the perspective of image super-resolution and image enhancement,this article conducts in-depth research on image super-resolution reconstruction methods,principles,and application fields such as feature exchan ge and texture migration,and conducts in-depth study on the principles and optimization methods of deep learning and generative confrontation networks.Based on a comprehensive analysis,the algorithm is effectively improved and optimized in view of the shortcomings and problems found in the actual application of several existing super-resolution reconstruction methods and generative confrontation network model.The main research contents and achievements of this paper in image super-resolution reconstruction are as follows:In order to further improve the accuracy of super-resolution reconstruction of low-resolution face images,so that the reconstructed super-resolution images have rich textures,a super-resolution reconstruction method for face images based on texture migration is proposed.This method uses the rich texture of the reference image to supplement the missing details in the low-resolution image,through an end-to-end model.First,the size of the low-resolution image and the reference image are scaled to the size of the target super-resolution image to be restored,and then the face image is feature extracted through the pre-trained Res Net network,and then the feature space is searched for matching the reference image Feature: Transfer the matched features to the super-resolution image in a multi-scale manner,and finally pass through a face structure constraint network to generate a face super-resolution image with richer texture details.This method makes full use of the effective texture provided by the reference image and the structural features of the human face,and focuses on the effective restoration of the texture of the facial features of the human face.Aiming at the problem that it is difficult to effectively restore the image details in the task of generating the confrontation network to perform the super-resolution of the face image,a method is proposed to improve the super-resolution reconstruction effect of the face image by combining the detail enhancement module.First,a generative confrontation network based on facial components is proposed,in which facial structure contours are restored by introducing a facial parsing module,and then a novel example-based detail enhancement algorithm is further proposed based on facial com ponent matching,and examples are matched by k-NN Image,which further solves the problem of insufficient image detail restoration.In view of the small and low-resolution face images,the current face recognition algorithm is not accurate,and a unified m odel is proposed.The face image is super-resolution reconstruction before face recognition,thereby improving the low Resolution accuracy of face recognition.First,through a super-resolution reconstruction model for face images,the high-frequency information of the face is initially restored.Later,a face recognition algorithm for low resolution is proposed,which combines residual network and image pyramid to recognize the reconstructed image.A large number of experiments show that this method is aimed at low-resolution face images,and its recognition effect is due to the current mainstream methods.