A method for blind motion blur removal of natural images based on l0 regularization

Abstract

The present invention provides a blind motion blur removal method for natural images based on L0 regularization, which uses the gradient information of the natural image and the sparse characteristics of the motion blur kernel to introduce corresponding L0 regular terms into the solution model of the blur kernel. When solving the model, first use the semi-quadratic splitting method to solve the intermediate restored image and blur kernel respectively. In order to increase the robustness, the pyramid model is used to solve the problem layer by layer in the solution process, and the estimation of the blur kernel only uses the gradient of the image. Information, through Fourier transform, the solution process is converted to the frequency domain, avoiding the direct deconvolution operation in the spatial domain, thereby achieving the purpose of rapid solution. Then, using the obtained blur kernel, the non-blind deconvolution method based on total variation difference is used to obtain the final restored image. The method provided by the invention has fast solution and high robustness, and the final restored image has good visual effects.

Description

Technical field [0001] The present invention relates to the technical field of image restoration, and more specifically, to a method for blindly removing motion blur from a natural image based on L0 regularization. Background technique [0002] Images are an important way for people to obtain information in modern society. However, in the imaging process, often due to camera shake or the movement of objects in the scene, the obtained image becomes blurred, which greatly hinders the normal use and subsequent processing of the image. [0003] Blind image deblurring can also be called blind image restoration or image blind deconvolution technology. Its purpose is to restore a clear image from a blurred image, and the blind removal of motion blur in an image is one of the important issues in image restoration. It is widely used in astronomical exploration, military reconnaissance and early warning, public security, and medical images. Processing, etc., has very important practical si...

AI Technical Summary

Problems solved by technology

[0004] There are two main difficulties in image blind motion blur removal. The first difficulty is how to accurately estimate the motion blur trajectory or also called blur kernel. Compared with non-blind image restoration, blind image restoration has higher pathological characteristics. Because it lacks prior knowledge and needs to estimate the two unknowns of blur kernel and clear image

The second difficulty is how to ensure that image blind motion blur removal has high robustness and achieves fast convergence

In the prior art, the prior constraints introduced in the simplest blind restoration model are based on the L2 or L1 paradigm. Although the introduction of the L2 paradigm can make the restoration model quickly solved, it will cause the estimated blur kernel to be dense, which will affect the final The restored image quality, and the restoration model based on the L1 paradigm will result in the estimation of discontinuous and noisy blur kernels, which will affect the final restored image quality

Images