基于多边缘信息融合的图像修复模型

Image Inpainting Model Based on Multi-Edge Information Fusion

提出了一种基于多边缘信息融合的深度学习图像修复模型MEC,将图像修复任务解耦为边缘信息修复任务和条件图像修复任务。首先,MEC通过边缘信息修复网络对缺失图像的多种边缘信息进行修复;之后,修复的边缘信息通过信息融合作为先验知识来辅助缺失图像的修复。实验结果表明,MEC的修复结果无论是在直观感受还是在量化指标上都比现有的深度学习图像修复模型有进一步的提升。

Proposed a deep learning image inpainting model based on multi-edge information fusion,named MEC,which decouples the image inpainting task into an edge information inpainting task and a conditional image inpainting task.MEC repairs various edge information of missing images through the edge inpainting network,and then the inpainted edge information assists the inpainting of missing images through information fusion as prior knowledge.Experimental results show that the inpainting results of MEC are further improved than those of the existing deep learning image inpainting models both in terms of intuitive feeling and quantitative indicators.