Research On Image Denoising Algorithm Based On Convolution Neural Network

Image is one of the most direct information people contact,but the image will be damaged by noise in the process of acquisition,compression and transmission,noise will lead to the subsequent image processing task to produce wrong diagnosis,so the study of noise removal has a very important significance.In the early days,there were many traditional methods to remove noise,and many classical filters,such as linear filter and nonlinear filter,appeared.Many of the classic and simple methods were effective at the time and could be used to meet the needs of the time at a lower cost.However,compared with traditional methods,the method based on deep learning can achieve higher PSNR and SSIM in image denoising and improve image quality to some extent.At present,the image denoising algorithm based on deep learning is the most researched one based on convolutional neural network.In this paper,convolutional neural network is used to study denoising.In order to enhance the denoising effect of the image and retain the details of the image,the work done in this paper is as follows:1.A hybrid sparse block and attention mechanism based image denoising algorithm is proposed to address the issue of Gaussian noise removal in images.This algorithm introduces deformable convolution and combines it with extended convolution and ordinary convolution to form a "hybrid sparse block" to complete the main denoising task.Adopting a dual attention mechanism allows the network to pay attention to more important information,thereby mining deeper levels of noise information;Design the network as a residual structure to alleviate problems such as vanishing gradients and slow training.The experimental results show that compared with other denoising algorithms,this algorithm has excellent denoising performance and can better preserve image details.It has higher data evaluation indicators and reflects better robustness.2.A GAN network-based algorithm for removing mixed noise in images is proposed.Although the most studied noise removal method is the removal of Gaussian noise,in reality,images often have more than one type of noise.Considering the complexity of removing mixed noise,this paper utilizes the characteristics of generative adversarial networks for denoising,And design the generation module network as a structure of upper and lower branches,expand the width of the network,avoid excessive depth of the network,and construct residual blocks in the generation module;And use an improved BRN algorithm to ensure that performance does not decrease when processing tasks in small batches;At the same time,a new compound loss function is constructed to help the network train efficiently.In summary,this article proposes a denoising algorithm based on generative adversarial networks for removing mixed noise composed of Gaussian noise and salt and pepper noise.Figure [25] Table [7] Reference [52]...