Non-uniform consistent blur removal method based on image region division

Abstract

The invention relates to a non-uniform fuzzy removal method based on image region division, and belongs to the technical field of image processing. Comprising the following steps: extracting a feature map beneficial to subsequent reconstruction of a clear image in a single-frame blurred image, inputting the feature map into a strong blurring detection module and a weak blurring detection module, and respectively outputting a detected strong blurring region attention map and a detected weak blurring region attention map; carrying out site-by-site point multiplication on the attention maps of the strong blurring region and the weak blurring region and the feature map, adding an input feature map, and extracting feature information of the strong blurring region and the weak blurring region which are divided according to image components on the original feature map; and respectively inputting the feature information into a decoder module for removing strong blurring and a decoder module for removing weak blurring, simultaneously reconstructing potential clear images by adopting two decoder branches, respectively obtaining the images after the strong blurring and the weak blurring are removed, and then inputting the images into a feature fusion module to generate a complete final clear image after the blurring is removed.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to a non-uniform and uniform blur removal method, in particular to a non-uniform and uniform blurred image deblurring method based on the guidance of image area division. Background technique [0002] Compared with the uniform uniform blur removal problem, the non-uniform uniform motion blur removal problem is more complex and difficult. The paper "Nah, S.; Hyun Kim, T.; and Mu Lee, K. 2017. Deep multiscale convolutionalneural network for dynamic scene deblurring. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 3883–3891." proposed a In this paper, a method for removing blurring of non-uniform blurred images is proposed. Convolutional Neural Networks (CNN) model is used to fit the degradation reconstruction process of non-uniform and uniform blurred images, and the multi-scale CNN architecture is used to simulate the optimization of tradition...

AI Technical Summary

Problems solved by technology

[0003] However, due to the different purposes of blur removal in different blurred areas, such as strong blurred areas need to remove obvious blur, smooth areas need to sharpen and enhance details, and textured areas need to maintain details, using a conventional unified network training strategy to learn a non-uniform motion deblurring model, It is almost impossible to achieve the above goals at the same time

The method used in the literature generally regards image blocks of different blur forms and degrees as consistent, and uses a unified training strategy to process different image regions, ignoring the differences in different blurred regions. This unified training strategy can easily lead to Model overfitting, most non-uniform and consistent blur removal methods, including literature methods, fail to effectively deal with different blur regions

Images