Perturbation method for restoring blurred images

Abstract

The invention discloses a perturbation method for restoring blurred images. The perturbation method for restoring the blurred images includes the steps: estimating a blurring model and blurring parameters of a to-be-restored blurred image by means of an existing blurring parameter estimation method, and constructing a blurring function h; perturbing each matrix value c of the to-be-restored blurred image by a perturbation trace d so as to obtain an error improvement value e of each perturbation; computing the range influenced by the maximum error improvement value e according to the blurring function h, and obtaining points of the influenced range; if the maximum error improvement value e is larger than zero, performing trace convergence for the points of the maximum error improvement value e so that pixel values of the points of the range influenced by the maximum error improvement value e are changed; and repeating the steps until the maximum error improvement value e is smaller than or equal to zero, and then terminating the algorithm so that the restored image is obtained. By the perturbation method, border effects can be eliminated to some degree, and direct processing of space variation during blurring can be facilitated.

Description

technical field [0001] The invention relates to an image deblurring algorithm, which belongs to the field of image processing. Background technique [0002] At present, in the process of image capturing, the captured images may be blurred due to inaccurate focusing, movement of the camera or the subject, and the like. And deblurring can improve the utilization quality of the image. [0003] The current image deblurring algorithms can be divided into two categories in the operational domain: frequency domain and spatial domain. [0004] Let the real scene be f, the blurred image obtained is g, g is the result of convolution by the degrading function h of f, plus the noise n, that is: g=f*h+n. [0005] Generally, considering that the blurring process of each place in the entire image is consistent, h is a function independent of pixel coordinates (x, y). In actual situations, the blurring process of different areas of the image may be quite different. [0006] The basic fre...

AI Technical Summary

Problems solved by technology

There are two situations in the frequency domain method that may make the restoration effect unsatisfactory: one situation is because the frequency domain transformation assumes that f is a periodically repeated image, and there will be a certain discrepancy with the actual situation in the boundary area, which will have a certain impact on the final result. Error, when the size of h is close to the size of g, the deblurring effect is poor; the second case is that if the division method is used to realize the fuzzy inverse transformation, the zero point in the frequency domain needs special treatment, such as the Wiener filter that introduces noise-signal ratio The estimated value of , eliminates the frequency domain zero point, but no matter whether the noise-to-signal ratio estimation is accurate or not, it will produce a certain degree of blurring on the restored image, that is, the real edge is blurred to a certain extent

Images