Research On Image Regularization Method Based On Structure Tensor And Total Variation Spectrum Response

In the process of people acquiring image information,there are many factors that affect the quality of the image,resulting in degraded image quality and loss of image information,which affects the visual effect and limits the analysis and processing of the latter image.Therefore,improving the image quality to make the image closer to the original image has become one of the hot topics in the field of image processing.Image restoration technology is occupying an increasingly important position in the field of image processing.On the one hand,this paper introduces the research status of image restoration technology and related basic theoretical knowledge,which focuses on the theory and research status of regularized image restoration model,and focuses on how to maintain the edge details of images more accurately and efficiently.And improve the quality of the image.In this paper,the linear integral convolution and the non-local structure tensor are combined to construct a new anisotropic non-local structure tensor,which can better obtain the local structural information of the image.At the same time,the regularization parameter based on local spectral response is proposed.The selection method constructs the regularization parameters with the defined local spectral response and applies them to the total variation model,so that the image detail information is better preserved.The main works of this Master's thesis are as follows:(1)In order to solve the current non-local structure tensor calculation method can not fully utilize the anisotropy of structural tensor,this paper proposes a new anisotropic non-local regularization method,which avoids the use of atomic decomposition strategy.Euclidean distance is used to evaluate the similarity process of tensor.At the same time,the method combines the atomic decomposition strategy with the extended line integral convolution using non-local mean filtering technique,and uses the tensor direction decomposition strategy without reconstruction to avoid the use of atomic decomposition.The reconstruction error caused by the strategy makes full use of the spatial correlation of the tensor for anisotropic smoothing.The numerical experiments show that the new anisotropic non-local structure tensor proposed in this paper can effectively extract the local structural information of the image and has good performance in corner detection.(2)In order to improve the image restoration effect of the total variogram image restoration model,this paper proposes a method for selecting the regularization parameters of the total variogram model based on the local spectral response.The total variogram transformation framework and the spectral response will be the total variogram restoration model.The spatial domain is extended to the frequency domain.This paper constructs the local spectral response to reflect the smoothing speed of the local detail in the image smoothing process of the total variation model,and then constructs the regularization parameter with the local spectral response,so that the regularization parameter can adapt to the image.The local structure information makes the total variation image restoration effect improve.Numerical experiments show that the regularization parameter selection method based on local spectral response makes the total variation image restoration significantly improved in visual perception and peak signal-to-noise ratio.