Research On Image Restoration Technology Of Transform Domain Based On Deep Learning

Images are one of the important ways for humans to obtain information.However,images are susceptible to interference during the acquisition or transmission process,which will affect the quality of image information.Image repair technology is the process of recovering lost or damaged images with algorithms.Image repair technology can be divided into application branches such as image super-resolution,image enhancement and image denoising according to the type of repair.Image denoising is the most effective way to improve image quality.Denoised images have richer features and give people a better visual experience.Therefore,image denoising technology is one of the research hotspots of scholars,and has been widely used in the fields of surveillance video,medical processing and satellite remote sensing.It is of great theoretical and practical value to study denoising algorithms with stronger denoising performance and higher ability to repair features.Therefore,this paper will study the denoising algorithms in image repair technology.This article first studies the basic principles of several commonly used image denoising algorithms,including traditional image denoising algorithms and deep learning-based denoising algorithms.In the traditional denoising algorithms,they can be divided into filtering algorithms based on the spatial domain,filtering algorithms based on the frequency domain,and algorithms that combine the filtering of the spatial and frequency domains.Among them,the filtering algorithms that combine the best performance of the spatial and frequency domains are the best.However,the denoising effect is unsatisfactory.The inability to distinguish between noise and details makes it difficult to effectively eliminate highintensity noise and repair details.With the rapid development of data,compared with traditional algorithms,the denoising algorithm based on deep learning has shown obvious advantages.It can automatically learn a large number of image feature information in CNN networks.By designing different network structures,it solves the problem of traditional algorithms that cannot distinguish between noise and details.In the algorithm based on deep learning,according to the noise model and type,different network structures are designed to accurately remove the noise.However,the deep learning network based on spatial information makes the performance of the denoising algorithm restricted to a certain extent,and it seems impossible to better adjust the denoising and repair details,which may cause the situation of residual noise.Aiming at the problems of existing algorithms,this paper proposes a SAR image denoising algorithm based on deep learning combined with spatial and frequency domains.Considering the complementarity between the wavelet sub-band features of the spatial structure features,the network is mainly based on wavelet domain information and supplemented by airspace information.The integration of airspace and wavelet domain information improves the learning ability of the network.The method restores image detail feature information to achieve image denoising.And by introducing spatial and channel attention modules,the network’s characterization ability is greatly improved,thereby improving the accuracy of denoising.According to a large number of experiments from the quantitative and qualitative perspectives,compared with the existing algorithms,the algorithm in this paper has better denoising effect and richer repair details.In addition,in order to prove the applicability of joint spatial and frequency domains to deep learning network denoising,a Gaussian noise denoising algorithm based on deep learning and joint spatial and frequency domains is also proposed.For Gaussian noise,the network structure is mainly based on spatial information and supplemented by wavelet domain information.Combining spatial and channel attention modules,the network learns noise and then uses the residual structure to achieve denoising.No matter from subjective vision or objective indicators,experiments show that the algorithm’s ability to denoise and repair detailed features is superior to the existing Gaussian denoising algorithms,which further demonstrates the effectiveness of the idea of joint space and transform domain.