Research On Image Denoising Based On Convolutional Neural Network

During the process of image acquisition and transmission,the imaging equipment is subject to external environmental interference,such as the interference of light intensity on the equipment,which will cause a certain noise in the acquired image band.The defects of the device itself or the relative motion between the device and the object will also pollute the images during acquisition and transmission.Noise is ubiquitous,and the images we see in life are basically polluted by noise.These noises will cause a certain interference to the image recognition,and the performance of the image processing algorithm will be degraded.Therefore,it is of great significance and practical value to carry out the research of image denoising.In this paper,the convolutional neural network is used as the research tool to study the task of By reasonably designing the networks,good performance of image denoising and fast computation speed are both achieved.The specific denoising content is summarized as follows:1.Image denoising network model based on multi-scale and residual learning is proposedTo better restore a clean image from a noise observation under high noise levels,we propose an Image denoising network model based on multi-scale and residual learning.Instead of using filters with different large sizes in traditional multi-scale schemes,we arrange multi-layer convolutions with the filters of the same size to speed up the model.And some dilated convolutions of different rates are combined with the common convolutions to enrich the extracted features in multi-layer convolutions.Furthermore,we cascade the multi-layer convolutions with residual blocks to improve the performance of image denoising.Our extensive evaluations on several challenging datasets demonstrate that our proposed model outperforms the state-of-art methods under all different noise levels in terms of PNSR,and the visual effects achieved by our proposed model are also better than the competing methods,and the proposed model has achieved a good denoising effect for high noise and Color image denoising.2.Image denoising network model based on cyclic residual block is proposedIn order to ensure the denoising performance and further reduce the running time of the model,an image denoising network model based on cyclic residual blocks is proposed.The network model introduces a cyclic structure,using the residual block as the basic model,and gradually removes the noise in the image through repeated iterations,reducing the number of network layers and parameters.In addition,the model uses a mixed loss function of L1 and L2,which can get better denoising effect.The experimental results on two datasets with different levels of noise show that the visual effect of denoising and the qualitative analysis of PSNR are better than the most advanced denoising algorithms at present,and the computation speed also exceeds that of most advanced image denoising algorithms.