Research On Image Restoration And Related Methods In Image Processing

Image restoration is one of the hot research issues in the field of image processing,and has a wide range of applications in the fields of computer vision,computer graphics,medical image processing,and remote sensing imaging.The original concept of image restoration refers to the use of known damaged image information to repair the destroyed region according to certain rules,as close as possible to or completely restore the original image.Later,with the deeper research,the concept of image restoration was extended,and ill-posed problems such as image super-resolution and denoising were introduced in the original field.Image smoothing is derived from image denoising,which aims to effectively remove texture details while preserving structural edges,thereby improving the effects of image edge detection,target recognition,and detail enhancementThe ill-posed problem has brought great challenges to the research of image restoration,and many scholars have made unremitting efforts for it.With the help of external image libraries,the quality of image restoration can be effectively improved,but it takes a very long time and can also cause artifacts and other issues.Making full use of the non-local self-similarity and local feature inforation of the image can effectively improve the effect of the algorithm while reducing the time complexity.Based on the idea of image segmentation,making full use of the self-similarity of images,weighting and fusing the repaired results,and continuously improving the image quality in iterative optimization.This paper focuses on the difficulties and deficiencies in image restoration and related problems.The main contents are as follows:1.Based on image self-similarity,combined with the idea of image segmentation,this paper proposes a cubic surface fitting image enlargement method with local high-frequency features as constraints.The cubic fitting surface constructed with high-frequency information and distance as constraints effectively improves the approximation precision and shape retention ability of the surface,and avoids the presence of sawtooth while maintaining the high frequency information of the enlarged image Double-filtering the initial enlarged image based on the self-similarity of the image can effectively further improve the quality of the enlarged image.Since the information in a given low-resolution image is not sufficient,there will inevitably be errors between the fitted surface and the real surface.Therefore,it is necessary to optimize the surface patch through continuous iteration.In order to speed up the convergence speed,a cubic surface with higher fitting precision is constructed for the error image.Since the method in this paper effectively maintains the local high-frequency feature information of the image,it not only has a higher PSNR and SSIM,but also has better visual effects at high-frequency information than other algorithms2.Based on the principle of image self-similarity and singular value decomposition,this paper proposes a novel iterative adaptive global denoising method.Based on the structural complexity of the image patch,the size of search window is adaptively determined.In order to reduce the interference of noise on the similarity between image patches,a multi-scale similarity measurement method is introduced.In order to ensure that all image patches are denoised,an adaptive step size and number of image patches iteration are proposed.Theoretically,the step size is changed to 1,thereby realizing the denoising of all image patches in different iterations.It not only guarantees the speed of the method,but also effectively reduces the artifacts.Based on the correlation between the singular value and the noise level,the denoised singular value is estimated,so as to realize the denoising of the noisy image patch matrix.The method in this paper has a higher PSNR and FSIM,and at the same time has a faster algorithm speed and good visual effects3.Based on the idea of image segmentation and decomposition,this paper proposes two two-stage image smoothing methods based on edge patch histogram equalization and patch decomposition.Both methods are divided into four parts as a whole.Firstly,divide the image into patches reasonably,and avoid dividing the same structural edge into different image patches as much as possible.Second,the image patches are divided into edge patch and non-edge patch based on the edge pixel ratio.The edge patch contains a large number of edges,and the histogram is used to equalize it,thereby increasing the gradient of edge pixels.In order to ensure the color consistency and continuity between image patches,the edge patch are inversely equalized.Third,decompose each image patch and extract smooth components,thereby reducing the gradient of the texture region.Fourth,each image patch and the entire image are sequentially smoothed with L0.The difference is that in order to divide the image more reasonably,the new method introduces the image energy map.Find the minimum energy seam in the designated region,so as to realize the non-linear and non-uniform image segmentation,which effectively reduces the seams caused by the segmentation and equalization.Combining the gradient direction difference between structural edges and texture details,the new method proposes L0 smoothing with gradient direction constraint,and at the same time uses the edge pixel ratio of the image patch as the weight of the relaxation factor.Compared with the existing technologies,the method in this paper is better in in maintaining the structure edge completely and effectively removing the texture details.Among them,the new method has better visual effects.