Research On Image Denoising And Super-Resolution Reconstruction Algorithm Based On Deep Learning

In recent years,with the rapid development of computer vision and deep learning technology,digital image online processing and information recognition technology are widely used in scientific research,military industry,medicine,smart city and other application fields on the premise of high-quality input images.However,in the process of image acquisition,processing,transmission,etc.affected by factors such as the environment,acquisition equipment,human beings,etc.the image will inevitably be defaced by different degrees of noise and exist in the form of lower resolution.Image denoising and image super-resolution reconstruction are classic and popular low-level vision research topics in the field of digital image restoration research.They can not only improve the visual perception quality of images,also improve the accuracy of subsequent high-level semantic analysis.At present,compared with traditional methods,the reconstructed images using deep learning methods have higher quality and better visual effects.Therefore,this thesis studies the image restoration algorithm based on deep learning,and mainly studies how to use the deep learning method to establish a more efficient algorithm model to solve the two problems of image denoising and image super-resolution reconstruction.The main research results of this thesis are as follows:(1)An image denoising model based on the improved U-Net neural network is proposed.This model is a denoising algorithm model designed based on the U-Net network structure,which consists of a denoising module and an edge information extraction module.The improvement of the network model is reflected in: First,the skip connection in U-Net++ is applied to the original U-shaped denoising subnet.The densely connected U-shaped denoising network can reduce the semantics between the encoder and decoder feature maps.Gap to restore a clearer image.Secondly,the edge information extraction module based on the VGG-16 network structure extracts the features of the image processed by the denoising network,and at the same time,the U-shaped denoising module is reversely optimized to restore a more real image.Experiments show that this thesis proposed algorithm is better than the current representative denoising algorithms in the objective evaluation index of the image when tested on common data sets and cross-type medical data sets,while the edge details and texture features of the image are better.Sharper,more realistic,and visually better.(2)An image super-resolution reconstruction model based on an improved generative adversarial network is proposed.Aiming at a series of problems in existing image super-resolution reconstruction algorithms,such as difficulty in network training,unclear generated images,and poor visual effects of generated images,based on the network model of generative adversarial network super-resolution algorithm(SRGAN),a network model was proposed.An improved image super-resolution reconstruction algorithm.First,the densely connected residual block is used to replace the original residual block in the generative network structure,and the batch normalization layer is removed to reduce the computational complexity.Second,the VGG-19 network is used as the basic architecture of the discriminant network,and average pooling is used to replace the original fully connected layer to prevent overfitting.In terms of loss function,the perceptual loss function,adversarial loss function and content loss function are introduced to form the overall objective function of the generator to jointly optimize the model.The improved algorithm adopts the Charbonnier loss as the content loss function to evaluate the similarity between the generated image and the real image and uses the WGAN-GP theory to optimize adversarial loss of the model to accelerate convergence.The experimental results show that the image generated by the improved algorithm is clearer,the texture details are more realistic and richer,the image can be synthesized better,and the overall visual effect of the reconstructed image can be improved.To sum up,this thesis conducts applied research on image restoration tasks using deep learning methods.Both denoising and image super-resolution reconstruction tasks show better visual effects.