Face Super-Resolution Reconstruction Based On Generative Adversarial Nets And Face Recognition

During recent years,face recognition has been a hot research topic in the field of image recognition,and has achieved rapid development,which has been widely used in many fields.Today,many security departments,sensitive public places,traffic arteries,residential quarters and so on are all equipped with real-time video surveillance systems.High-quality input images have become the prerequisites for most face recognition models to achieve high performance.However,due to factors such as the resolution of the surveillance camera,the lighting conditions of the surveillance environment and the target distance,the video images captured by the surveillance system may have low quality.The low resolution of the image and missing facial detail affects the further face identification.Therefore,how to improve image quality and study efficient super-resolution reconstruction technology becomes crucial.In order to solve this problem,this thesis explores face super-resolution technology using deep learning and builds a high-performance reconstruction model.Then,a preliminary study on the application of the technology in face recognition is made.The research content of this paper is:1.The typical image super-resolution reconstruction algorithm is analyzed.The advantages of convolutional neural network algorithm in image reconstruction are proved theoretically.This thesis proposed the super-resolution reconstruction algorithm to maintain human face characteristics.In this paper,when constructing the reconstruction network,we select the appropriate parameters of the convolution network for the image with distinctive features such as face and use the improved loss function.By this method,it is possible to maximize the reduction of detail features based on the invariance of the overall facial features.2.A super-resolution reconstruction algorithm based on generative adversarial networks is improved,which makes the effect of reconstructing the face image further improved.As a new training method,this paper mainly focuses on the reconstruction of face image from three aspects of network construction,objective function and training set.The generation network enhances the edge contour of the face based on the improved Laplacian pyramid,making the image more natural with multi-level feature loss.In addition,this paper constructs a multi-scale,multi-scene,multi-pose training set,which improves the robustness of the algorithm to different real scenes.3.The entire face recognition framework was built,and the application of super-resolution reconstruction algorithm in actual scenes was explored.Face detection,face normalization,feature extraction and face matching are implemented,and the algorithm of face super-resolution reconstruction is embedded in it.The test results show that the reconstruction algorithm proposed in this paper can effectively improve the effect of super-resolution reconstruction of face images and can improve the accuracy of low-resolution face recognition by nearly 15%.