CrackSegConnect: a Crack Inpainting Networkbased on Segmentation Model

Crack is one of the most common structure distresses which often appear in the engineering construction systems, and thus crack detection and evaluation is particularly important in structural health monitoring. However, defects in the captured crack image occur frequently due to the complex circumstance conditions, bringing difficulty to identification work and observation of crack property. Therefore, it is necessary to inpaint the crack before detection and evaluation. In this paper, we propose a crack inpainting network based on segmentation model. The network introduces segmentation results as prior knowledge, which can improve the inpainting results. The experimental results have demonstrated that the proposed network can inpaint the crack structure with high accuracy.