Research On Fast Super-resolution Reconstruction Of Noisy Images

With the rapid development of 5G communication technology,high-broadband and low-latency bring convenience,but also greatly stimulate the demand for highdefinition images.In many digital image application fields,such as social security monitoring and medical image segmentation,rich image details are required to provide more effective information.However,in practical applications,a large number of uncertain factors such as the capability of the imaging equipment and the aliasing in the frequency domain during transmission,or even the relative displacement of the object and the camera during the shooting process.In order to solve this problem,it was decided to use image super-resolution The reconstruction technique aims to reconstruct a compressed image from one or more low-resolution images input.The process of degrading high-resolution images to low-resolution images is accompanied by a large amount of image edge details and high-frequency information loss,so the reconstruction process is a typical morbid problem.The unknown noise during shooting and storage makes the reconstruction process more difficult.In order to obtain more unique quasi-solutions close to real images,this paper studies the problem of superresolution reconstruction of noisy images,and proposes a fast and efficient superresolution reconstruction algorithm.The main research results include:(1)Most super-resolution reconstruction algorithms now assume that the input image is noise-free.However,in the actual application environment,low-resolution images with noise are very common.Aiming at the problem that current algorithms have limited ability to reconstruct noisy images,a fast super-resolution method with obvious suppression effect on noisy images is proposed Reconstruction algorithm.(2)Unlike traditional algorithms,this method completes super-resolution reconstruction of the image while processing noise.Among them,in the training phase of the dictionary,the traditional training method is borrowed and the texture structure of the low-resolution image block is added.It is worth noting that the example images used in training are noise-free.So even if the noise amplitude of the input image changes,there is no need to retrain the dictionary.(3)In the reconstruction process,the column atoms of the sparse dictionary are used as matching objects,on the one hand,the calculation cost of finding similar patches from millions of matching objects is reduced.On the other hand,sparse representation can concentrate signal energy on individual atoms and suppress noise interference.Then,the normalization processing of the input feature vector and the sparse dictionary improves the accuracy of the matching,and combines the similarity matching and weight limit model to complete the weight distribution of similar blocks,and reconstruct the corresponding high-resolution image blocks..Finally,the estimated high-resolution image and denoised low-resolution image are obtained by weighted average.The two reconstructed images are combined with iterative back projection to obtain the final estimated high-resolution image.The algorithm of this paper is verified in the routine image test set,and the results are better than the comparison algorithm and have better robustness to the simulated Gaussian noise.