In order to effectively fill the holes in the depth map, especially to repair large holes, this paper proposes a new information entropy-guided strategy for depth inpainting. Firstly, the depth map and color image undergo preprocessing to obtain a connected hole map and grayscale image for subsequent steps. The filling priority of invalid points is then evaluated by introducing the concept of information entropy, which improves the intelligence of the filling priority evaluation results. Secondly, the depth value prediction for invalid points is guided by color and gradient information, which ensures the integrity of the overall depth data and enhances the precision of the restoration. Finally, a comparison experiment was conducted on the Middlebury dataset. The proposed method demonstrated better robustness and accuracy compared to other competing methods, providing new ideas for related research in the field of deep image processing.