<p>In view of the style characteristics of traditional art landscape painting, we propose an inpainting method based on structural information that imitates the restoration steps of traditional masters. The method involves first outlining the damaged edge structure and then filling in the texture. In this paper, we present a novel architecture called Structure-Guided Multi-Scale Network (SGMS-Net) that is specially designed for art image inpainting. To extract structural data, we use a convolutional neural network-based HED network and input it along with the missing image. To capture local information at different semantic levels, features are extracted by each multi-scale focal layer in both the structural channel and texture channel. Moreover, AOT-Block fusion is used to collect global consistency features from the remote background region and missing image context for context reasoning. The experimental results demonstrate that our proposed inpainting algorithm has higher processing efficiency, produces more realistic visual results, and yields a more reasonable structure.</p>