A Non-Blind Deblurring Method for Blurred Images Based on Adaptive Gradient Sparse Model

Abstract

The invention discloses a non-blind deblurring method for a blurred image based on an adaptive gradient sparse model. The non-blind deblurring method does not adopt a fixed shape parameter value or a fixed scale parameter value in gradient distribution estimation for an image, but adopts different shape parameter values and scale parameter values in allusion to different pixels, so that the method is enabled to well adapt to variations in image texture, and thus a restored image restored by using the method disclosed by the invention is enabled to have a high signal-to-noise ratio. Meanwhile, on the aspect of the subjective quality, a smooth region of the restored image does not have a noise point and appears smooth in a natural manner, a texture region of the restored image is clearer, and more excellent subjective visual quality is acquired. The image is divided into the smooth region and the texture region, a fixed shape parameter and a fixed scale parameter are directly adopted for pixels belonging to the smooth region, a global convergence algorithm is adopted for pixels belonging to the texture region to estimate the shape parameter and the scale parameter, and a better estimation result can be acquired under less data.

Description

technical field [0001] The invention relates to an image processing technology, in particular to a non-blind deblurring method for blurred images based on an adaptive gradient sparse model. Background technique [0002] In the process of image acquisition, transmission and storage, due to the influence of various uncertain factors, such as the relative shake of the aircraft imaging equipment, inaccurate camera focus, image compression coding, etc., as well as the noise introduced in the process of shooting, storage and transmission etc. will cause image quality degradation, resulting in blurred images acquired by the receiving end. Image restoration is to remove or reduce the quality degradation phenomenon in the process of image acquisition, transmission and storage, and restore the characteristics of the original image as much as possible. Among them, the restoration of blurred images is also called blurred image deblurring, which is to restore the original information of...

AI Technical Summary

Problems solved by technology

[0003] At present, the researches on non-blind deblurring of blurred images mainly include: 1) Deblurring of blurred images based on Wiener filtering. Deblurring, the method has the advantages of simple principle and high computational efficiency, but the disadvantage is that it needs to know the power spectrum of the noise and the original clear image, which is often impossible in practical applications, and the restored image (the image after deblurring ) will have ripple-like ringing distortion around strong edges

2) Use the Lucy-Richardson (LR) method to deblur the blurred image. This method assumes that the blurred image obeys the Poission distribution, performs maximum likelihood estimation modeling under the probability framework, and uses the alternate iteration method to solve the restored image. This method cannot Given specific termination iteration conditions, when the number of iterations exceeds a certain value, the ringing phenomenon and the influence of noise will be aggravated

3) Deblurring the blurred image based on the regularization method. Since the regularization term is quadratic, the calculation is simple, but the restored image will be over-smoothed

4) The image gradient sparse prior model based on the prior information of the image deblurs the blurred image. Since the blurred image is generally stable in blocks, the local features of different texture regions are obviously different, so a single method is used for the entire blurred image. Image gradient sparse prior models are disadvantageous for accurate reconstruction of local information

5) Different gradient priors are used for textured areas with different content in a blurred image. Although the visual quality in the textured area has been improved, it is accompanied by more distortion and noise in the smooth area.

This method uses gradient sparse priors in the entire image, so this method can remove ringing distortion and noise, but it also makes the result of deblurring show piecewise smoothness, loses intermediate frequency texture information, and leads to visual quality On the other hand, this method often uses a fixed λ value and p value for an image, but the image is generally block-stationary, and the local features of different regions are obviously different, so a single p-value regularization constraint is used The model is not good for detailed reconstruction of local texture information

Images