| As the visual foundation for human perception of the world,images are essential for acquiring and transmitting the information.However,images are inevitably affected by noise interference during the collection and transmission process,which can decrease image quality and affect subsequent analysis and processing.With the continuous development of computer technology,image preprocessing technology has been widely used,and image denoising technology,as an important part of image preprocessing,has essential research significance.This article uses deep learning models for image denoising,Traditional deep learning denoising models often lead to blurred image details and a lack of authenticity in image edges after denoising.To solve these problems,this paper studies the image denoising method based on deep learning to further improve the problem that the traditional model is easy to lose image details in the process of denoising.The main contributions of this paper are as follows:Firstly,an image denoising model MR-GAN(Multiscale module and Residual structure-Generative Adversarial Networks)based on GAN is proposed in this paper.The model utilizes multiscale modules to extract multidimensional features at different levels,thus better capturing the details and texture information in the image.At the same time,a residual structure is adopted to avoid losing too much image detail during the denoising process.The experimental results show that the image quality is significantly improved after denoising by MR-GAN for noisy images.Secondly,in order to improve the instability of traditional GAN networks during adversarial training and to solve the problem of traditional CNN(Convolutional Neural Networks)unable to distinguish key features suitable for restoring images from redundant and irrelevant features,an image denoising model RMA-WGAN(Residual structure and Mixed Attention-Wasserstein Generative Adversarial Networks)based on Wasserstein GAN is proposed.RMA-WGAN uses an improved composite loss function to guide network parameter optimization.Through experimental comparison and verification,the training process of RMA-WGAN is more stable than that of traditional GAN networks,and the denoised images are closer to real images.Experimental results show that RMA-WGAN has better denoising performance than other comparison algorithms,The average PSNR(Peak Signal to Noise Ratio)value of the images before and after denoising is increased from 24.71 d B to 33.76 d B,an increase of 36.6%,and the average SSIM(Structural Similarity Index Measure)value is increased from 0.8451 to 0.9503,an increase of 12.4%.The visual effect is significantly improved after RMA-WGAN denoises the noisy image.Finally,according to the requirement of image denoising,this paper designs and develops an image denoising system based on RMA-WGAN.This system can directly output a clean image after denoising by inputting an image with noise interference,making it a visual image denoising tool.After using this system,the quality of the denoised image meets the requirements of practical applications. |
