Dunhuang murals, treasured elements of our cultural heritage, have faced significant degradation due to prolonged environmental exposure, emphasizing an urgent need for sophisticated restoration methods. Current inpainting techniques, while valuable, often compromise the intricate balance between a mural's overall structure and its minute texture, leading to subpar restorations with blurred regions and inconsistencies. To address these shortcomings, our research introduces the Multi-Stage Progressive Reasoning Network (MPR-Net). Drawing inspiration from traditional manual restoration approaches, the MPR-Net integrates two specialized networks: the Structure Feature Reasoning Network (SFR) for primary structural restoration, and the Texture Feature Reasoning Network (TFR) for enhancing and refining the mural's detailed textures. Ensuring a seamless fusion of features across varying scales, our model also incorporates a Multi-Scale Feature Aggregation Module (MFA), optimizing the accuracy and richness of the inpainting process. Compared with other state-of-the-art methods, the effectiveness of the proposed method in image restoration is verified both quantitatively and qualitatively.