Dynamic interactive holography with semantic-aware occlusion modeling from a single RGB image

In recent years, computer-generated holography has demonstrated significant potential and research value in emerging display technologies such as virtual reality and augmented reality, owing to its unique advantages in 3D holographic reconstruction and algorithm-driven encoding. However, existing hologram generation methods based on single-view images are primarily designed for ideal Lambertian surfaces, making it difficult to accurately model complex light field behaviors in non-Lambertian materials, such as specular reflections and virtual images. Additionally, these methods often fail to effectively extract the implicit 3D structural information from images, thereby limiting their interactive capabilities. To address these challenges, we propose a semantic-aware interactive holography that generates editable 3D point cloud holograms from a single RGB image, supporting complex occlusion modeling and dynamic interaction. Numerical simulations and optical experiments demonstrate that the proposed method accurately estimates depth in non-Lambertian scenarios such as mirror reflections and underwater objects, enabling dynamic editing and display, and significantly enhancing the realism and interactivity of holographic displays.