Research On Anti-Noise Image Super Resolution Algorithm Based On Deep Learning

Images are a simple medium for conveying information quickly.However,due to the influence of image acquisition equipment,environmental noise,information transmission requirements and other factors,usually only low quality images,that is,low resolution images.Super resolution reconstruction technology based on deep learning can improve image resolution by means of software and algorithm,so as to be widely used in the fields of medical treatment,intelligent display,monitoring and remote sensing image.This kind of technology has great development potential and can bring more opportunities and challenges to engineering control and other fields.Nowadays,the continuous improvement of computer hardware computing ability drives the continuous development of artificial intelligence technology,and deep learning has achieved excellent results in the field of computer vision.At present,image super resolution technology adopts deep learning method for research,which usually requires the use of a large number of samples for network training,so as to directly learn the mapping relationship between low resolution image and high resolution image.However,most of the current super-resolution reconstruction networks assume that the input low-resolution image is obtained through the ideal bicubic interpolation downsampling,while in reality the lowresolution image usually has unknown noise.This degradation mismatch makes most networks not applicable in the real scene.By fusing image noise reduction module into image super resolution reconstruction network,two algorithms are proposed in this paper to achieve clear super resolution images output by image super resolution reconstruction algorithm under realistic environment.The methods are as follows:(1)In this paper,a super-resolution reconstruction algorithm based on Laplacian pyramid network fusion denoising module is proposed.In this method,Denoising Module is fused on the basis of Laplacian Pyramid network(Lap SRN),so that the denoising module can directly reduce the noise of each middle layer image in super resolution reconstruction.In addition,recursive module is used to reduce the number of parameters in the network model,and introduces TVLoss and noise reduction loss.To improve the clarity of lowresolution image reconstructed by super-resolution network for unknown environment.In order to verify the effectiveness of the proposed method,the proposed method is compared with other advanced methods on existing benchmark datasets.Comprehensive experiments show that the proposed method can better reconstruct the image super-resolution in the real environment.(2)In order to realize the same function of LFDSR network,solve the problem of insufficient generalization and robustness of LFDSR network,and realize that the image super-resolution reconstruction network can output clear super-resolution images with different levels of noise in low resolution images,this paper proposes a noise-resistant multistep image super-resolution reconstruction algorithm.First of all,to combine information distilling image noise reduction network generated against network to network training,in order to improve the noise reduction network to contain different degree of noise image noise reduction ability,and secondly,characteristics of pure noise control network in the middle of the network diagram and noise image and step by step image super-resolution reconstruction after network combination.Finally,with the stepwise network training,the network can effectively reconstruct the low-resolution image in real environment.Experimental results show that the proposed method is better than other advanced methods for super-resolution reconstruction of low-resolution images with different levels of noise.