Research And Implementation Of Image Denoising Algorithm Based On Convolutional Neural Network

Image denoising has always been one of the hot research problems in the field of computer vision.The main purpose of image denoising is to remove noise from noisy images and restore potential noise-free images.Despite the rapid advances in camera technology over the past few decades,a large number of noisy images are still produced.Not only does this result in unpleasant visual effects,but it can seriously degrade the performance of visual systems,such as surveillance systems and traffic systems.In addition,image denoising is also the basis for other advanced computer vision tasks,such as object detection,face recognition,and semantic segmentation.In order to obtain more accurate denoising results,researchers have conducted in-depth research on this problem and proposed a large number of efficient methods.The main research contents of this paper are as follows:(1)Multi-scale attention interactive image denoising algorithm.Aiming at the problems of incomplete image detail feature extraction and low feature utilization by denoising model,a multi-scale attention interactive image denoising algorithm based on convolutional neural network is proposed.First,a multi-scale feature extraction block is used to extract the rich detailed features and spatial context features in the shallow network,and the features are fused.Then,a two-branches attention mechanism is used to enhance useful image feature information,suppressing useless noise while reducing the loss of key information.Finally,the paired convolution operations in the residual block are used to extract the multi-level local features of the network,and the dense connections are used to jointly learn the global features,which enhances the interactivity and feature reuse of the model,and effectively improves the denoising performance of the network.Experiments show that the proposed denoising network has improved denoising effect on three types of noise image datasets,namely gray-scale synthetic noise,color synthetic noise and real noise.(2)A dual attention supervised image denoising algorithm based on U-Net.Aiming at the problems of loss of feature information during U-Net downsampling and low feature utilization during upsampling,an image denoising algorithm based on U-Net improved dual attention supervision is proposed.First,in the encoding process,for the image features obtained by using the two downsampling methods,the dual-attention feature selection module will adaptively select the features in the channel dimension and the space dimension respectively,and select the more expressive complementary features for fusion.fusion.Then,in the decoding process,the feature enhancement module uses the fusion features generated in the encoding process to supervise the deep feature information in the decoding process,so as to enhance the key information,further optimize the image features,and improve the performance of the denoising network.The experimental results show that the proposed denoising network achieves significant improvement in denoising effect on three types of noisy image datasets including grayscale,color synthetic noise and real noise.