An image denoising method based on high-order overlapping group sparse total variation

Abstract

The present invention proposes an image denoising method based on high-order overlapping group sparse total variation. On the basis of the traditional first-order overlapping group sparse total variation technology, an intersection method based on high-order regular term constraints is proposed. Stacked sparse total variation image restoration method. The first-order overlapping combination technology generalizes the conventional total variation gradient of each pixel to a combined gradient, thereby improving the difference between smooth areas and edge areas, while the high-order regularization term derives from second-order or higher-order gradient information. Starting from this, the "staircase effect" can be alleviated more effectively, thereby improving the protection of image edges. In order to improve the operation speed of image restoration, we model the horizontal and vertical difference matrix operations of the image as convolution operations, combined with periodic boundary conditions, so as to cleverly apply the two-dimensional fast Fourier transform to the image restoration problem, using frequency Dot multiplication operations on the domain replace large matrix operations on the spatial domain.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an image denoising method based on high-order overlapping group sparse total variation. Background technique [0002] Since the Total Variation (TV) model was proposed, it has been widely used in various image processing problems such as image denoising because of its good processing effect. However, this model also has obvious shortcomings, that is, the "ladder effect" is relatively serious. [0003] Since the proposal of the total variational denoising model, various improved models aimed at removing its "step effect" have been proposed successively, such as anisotropic total variation (anisotropic TV, ATV), fractional total variation (fractional order TV, FTV), nonlocal TV (NLTV), overlapping groups parsity TV (OGSTV), and total generalized variation (TGV) models. [0004] Traditional image denoising methods usually use first-order anisotropic total variation methods. Compar...

AI Technical Summary

Problems solved by technology

Compared with the basic full variational model, although various improved full variational models have some improvements in overcoming the "step effect", there is still room for improvement.

Images