Repair Of Distorted Images Based On Deep Cyclic Convolution Neural Networks

With the passage of time,the development of digital information technology is changing rapidly,and computer technology and Internet resource sharing have been increasingly widely used in daily life.With the advent of the era of big data technology,people are increasingly dependent on information interaction.In addition,people have also put forward higher requirements for signal and information processing,mainly including image processing.Currently,due to various reasons such as poor hardware equipment and poor transmission environment,image information is lost during image acquisition,transfer,and transmission,resulting in varying degrees of image distortion,which greatly limits the quality,safety,and stability of work.This requires us to take some measures to repair these distorted images.Nowadays,many image repair technologies have achieved very good results,among which the application of deep learning in the field of image repair has become more popular and the repair effect is also getting better.Therefore,image repair based on neural networks has become a research hotspot at this stage.This is due to the particularly good self-learning ability of deep neural networks.The deep neural network learns from the training data in the dataset to obtain information about the internal structure of the image,and then summarizes the information to be used to achieve certain functions.In addition,neural network based methods can also be used in other areas of image processing,such as image super-resolution reconstruction and image denoising.In this paper,the relevant methods of image repair and super-resolution reconstruction using cyclic convolutional neural networks are used as the starting point.In order to further improve the quality of image repair,in-depth research has been conducted on issues such as edge blur and semantic information acquisition,and effective solutions have been proposed during the experimental process.The main research content of this article is as follows:(1)In order to improve the training speed of neural networks,this paper designs a Recurrent Convolutional Neural Network(RCNN)model.Firstly,the convolution module in feature extraction is reconstructed,and the pooled layer and fully connected layer that affect the efficiency and computing speed of feature extraction are discarded,reshaped into the form of Conv+BN+Re LU.The sample normalization layer is introduced and the nonlinear expression of the feature extraction module is improved.Then,a cyclic layer is introduced into the original shallow convolutional network model,and an end-to-end deep cyclic convolutional neural network is trained to ensure that the number of convolutional layers remains unchanged,while also increasing the depth of the entire network model.Finally,the final repair result is obtained through the restoration and reconstruction of the U-net network structure.The network model is analyzed in detail and verified through a large number of experiments.(2)In response to the current situation that people regard objective data as the only indicator for evaluating the quality of image restoration,this paper applies the Recurrent Convolutional Neural Network(RCNN)model to super-resolution reconstructed images,and conducts comparative experiments with the traditional Super-Resolution Convolutional Neural Network(SRCNN)model for image restoration.Based on the improvement of objective evaluation indicators,Improve the perception of images by the human eye.Experimental results show that the network model designed in this paper has better repair effects compared to traditional algorithms.