Motion blur restoration method based on contour enhancement strategy

Abstract

The invention discloses a motion blur restoration method based on a contour enhancement strategy, and relates to the technical field of image processing, and the method comprises the following steps:(1) encoding an original blur image, removing noise information through fine adjustment of a restoration network, and finally restoring a clear contour image through a decoder; respectively carrying out contour extraction on the image by using a Sober operator and a Canny operator, forming a sequence with the contour recovered by the decoder, carrying out further restoration extraction on the image contour by using LSTM, and finally generating a sharp edge; (2) respectively sampling and encoding the original fuzzy graph and the sharp edge, pairing the graph codes and the sharp edge codes withthe same size one by one, and outputting graph code and sharp edge code pairs; (3) generating a potential clear graph by using a multi-scale repair framework; according to the invention, the sharp edge in the image with serious motion blur can be extracted, so that the generated sharp edge effectively assists a multi-scale framework in removing the motion blur, and the efficiency of removing the motion blur is effectively improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a motion blur repair method based on an outline enhancement strategy. Background technique [0002] There is a problem of unclear edges in image blur restoration. In order to solve this problem, researchers have used various methods, including traditional methods and deep learning methods. [0003] In the restoration strategy through traditional methods, in order to restore images with clear outlines, researchers have made various attempts; Zhou Y, Komodakis N and Liang Chen et al. based on the maximum a posteriori estimation of the image Repair; Liuge Yang, Hui Ji et al. Edge selection for motion blur by maximizing variational expectation; Chang C, WuJ, Chen K et al. Removal of motion blur by contour restoration and bilateral filtering; however they are either for non-Gaussian noise Fixes are either poor or don't handle motion blur on complex tracks. Other scholars re...

AI Technical Summary

Problems solved by technology

[0003] In the restoration strategy through traditional methods, in order to restore images with clear outlines, researchers have made various attempts; Zhou Y, Komodakis N and Liang Chen et al. based on the maximum a posteriori estimation of the image Repair; Liuge Yang, Hui Ji et al. Edge selection for motion blur by maximizing variational expectation; Chang C, WuJ, Chen K et al. Removal of motion blur by contour restoration and bilateral filtering; however they are either for non-Gaussian noise The fixes are either suboptimal or don't handle motion blur on complex tracks

Other scholars repaired the image based on the estimation of the blur kernel. In order to obtain an accurate blur kernel, Cai C, Meng H, Zhu Q et al. divided the edge information of the image into strong edge and weak edge, and they used the existing edge The detection technology obtains the strong edge of the image, and uses the method of three filters to suppress the noise to deal with the weak edge part, but the effect of this framework is poor in the non-uniform blur scene; Yue T et al. improved the fuzzy kernel power spectrum estimation method, thus Eliminates the negative impact of structural image edges, and improves the accuracy of blur kernel estimation through a hybrid kernel estimation method that effectively fuses edge and power spectrum information, thereby improving the effect of image blur repair; however, not all blur kernels are reversible Therefore, the effect of removing motion blur in the image by the method of kernel estimation is unstable

[0004] With the emergence of deep learning, some researchers have applied neural networks to blurred repair scenes that need to improve the sharpness of the outline and achieved good results. Some researchers have added penalties related to the degree of edge sharpening in the loss function, such as Gradient constraints; however, directly adding a specific loss function will lead to unstable content restoration, because the content loss and edge loss will interact with each other in the training network, resulting in superimposed errors that may not be as good as unilaterally considered error training effects , thereby reducing the training effect; in order to solve this problem, S.Zheng, Z.Zhu et al. proposed that the contour of the image can be repaired first, they used the Canny operator and the Sober operator to extract the contour from the fuzzy image, and then based on the clear The contour of the image repairs the edge information of the image. Although their framework has achieved excellent results in the scene of removing Gaussian noise, it cannot directly use the method proposed in the paper for contour generation in the scene of complex trajectory motion blur, because Contour information and noise signal are mixed seriously, so it is impossible to extract enough contour information directly using the existing contour extraction operator

Images