Blurred image blind restoration method based on non-local window gradient

Abstract

The invention relates to a blurred image blind restoration method based on a non-local window gradient and belongs to the field of image processing. The method comprises the following steps that NLWGof a blurred image is smaller than the NLWG of a clear image, and the NLWG is suitable for all image blocks in the image; whether the image is a clear image or a blurred image, the pixel values of allthe pixel points can be normalized to be between [0, 1], and non-local window gradient prior is defined; a blurred image blind restoration model is defined based on non-local window gradient prior; optimal solution is carried out on the proposed model; and a final clear restored image is acquired by adopting a non-blind restoration method. A more accurate blurring kernel can be estimated, so thata higher-quality clear image can be restored.

Description

technical field [0001] The invention belongs to the field of image processing, and relates to a method for blind restoration of fuzzy images based on non-local window gradients. Background technique [0002] The success of existing blurred image blind restoration methods is mainly attributed to the proposal of various image priors, or edge prediction strategies. However, methods based on edge prediction, bilateral filtering, two-stage kernel estimation, block prior, color line prior, hyper-Laplacian prior, usually employ heuristic edge selection when strong edges do not exist in the blurred image , the image quality restored by these methods will have a significant decline. To avoid such heuristic edge selection, many algorithms based on natural image priors have been proposed, including normalized sparse priors, gradient priors, dark channel priors, and extremum channel priors, etc. These images perform well on general natural images, but cannot generalize well to specifi...

AI Technical Summary

Problems solved by technology

These images perform well on general natural images, but do not generalize well to specific scenes

Although dark channel priors can also be applied to these specific scenarios, intermediate latent image estimation methods based on dark channel priors are likely to fail when the image contains a large number of bright pixels

Although the extreme value channel prior is optimized on this basis, if there is a lack of a large number of bright and dark pixel values, the effect of this method will also be limited

Images