Image Smoothing Based On Global Structure Sparsity And Parameter Adaptation

Image smoothing is a basic image processing technology.The smoothed image often has less complex texture and highlights the main structure of the image.Image smoothing technology has a wide range of applications in computer graphics and computer vision.For example,image segmentation,edge extraction,image enhancement,image decomposition and artifact removal can be preprocessed by image smoothing,so as to facilitate the further operation of the image.The edge of image is the key information needs to be preserved in the process of image smoothing.Whether the edge of image can be protected more effectively is one of the decisive factors to evaluate the effect of image smoothing.There are three kinds of smoothing algorithms:the algorithm based on local information filtering is simple and fast,but there will be pseudo boundary;the algorithm based on global optimization is the current mainstream algorithm,but the idea of global iteration will destroy the weak edge information of image;the method based on data-driven improves the quality of smoothing results to a certain extent,but the universality is poor.At present,the most commonly used algorithm is still based on global optimization.The main problem of this kind of algorithm is that it is easy to lose the weak edges in the process of smoothing.In order to protect the weak edges effectively,this paper proposes a total variational image smoothing algorithm based on global sparse structure and parameter fitting.Total variation smoothing is a commonly used heuristic smoothing algorithm,which can get better result image by reasonably describing the target.Firstly,based on the global sparse structure,our algorithm decomposes the image into high frequency part and low frequency part.The low frequency part contains less texture information and is easy to be smoothed,while the high frequency part is more sensitive to edge information,so it is more suitable for the selection of smoothing parameters.Therefore,the algorithm calculates the smoothing parameters through the high frequency,and uses this parameter to smooth the low frequency.After that,the smoothing parameters are adaptively processed to ensure the local weak edge information of the image.Ideally,the algorithm needs to calculate the smoothing parameters of all pixels in the image,which is quite time-consuming.In order to reduce the computational complexity,the algorithm first calculates the sample value of the smoothing parameters in patches on the high frequency,then uses the Bessel method to fit the sample value into the parametric surface,and obtains the corresponding smoothing parameters according to the coordinate position of the pixel points.In the process of fitting,the algorithm improves the Bessel method based on Euclidean distance and pixel difference to ensure the accuracy of smoothing parameters.Compared with the traditional smoothing algorithm based on global optimization,the smoothing parameters given by our algorithm can adjust the local information of image more effectively.In order to solve the problem efficiently,we use Alternating Direction Method of Multipliers(ADMM)to unify the whole algorithm and obtain the smoothed results by iterative optimization.The adaptive smoothing parameters proposed in this paper can effectively adjust the local information of the image,and sample the parameters and fit the parametric surface,which provides a new idea for the adaptive processing of the total variation algorithm.The effectiveness of the proposed algorithm is verified by experiments.Compared with the traditional method and the machine learning method,it is proved that the algorithm can retain the weak edges better.In addition,the performance of this algorithm in image processing tasks such as edge extraction,image abstraction,pseudo boundary removal,image enhancement and content-aware image manipulation,which proves the effectiveness of this algorithm.This paper also discusses the parameter setting of the algorithm and gives the method of selecting the optimal parameters.