An Efficient Diffusion Depth Image Inpainting Method Based on RGB-Guided

Depth images play a crucial role in fields such as 3D reconstruction, simultaneous localization and mapping (SLAM), and virtual reality. However, depth sensors are often affected by environmental factors when capturing depth images, resulting in obtained depth maps that often contain areas of holes where information is lost, especially at object edge locations. The traditional inpainting method has achieved better results in depth image inpainting, but for holes that are in the edge position, the edges of the repaired image objects are prone to distortion and blurring; when the holes are large in size, the repair is incomplete or more costly in terms of time. In this paper, we propose an efficient diffusion depth image inpainting method based on RGB-guided. Firstly, the color image edges are extracted as depth-assisted edges to help determine the edge hole locations; then, the integrity of the depth image edges is restored by the local maximum filling method; finally, the improved curvature-driven diffusion model is combined to achieve accurate filling of larger holes. To validate our algorithm and compare it with other existing methods, we perform experiments on the Kinect dataset and the RGBZ dataset, providing qualitative and quantitative results, respectively. The results show that the method is effective and superior to other methods.