Automatic Texturing without Illumination Artifacts from In-Hand Scanning Data Flow
Résumé
This paper shows how to improve the results of a 3D scanning system to allow to better fit the requirements of the Multi-Media and Cultural Heritage domains. A real-time in-hand scanning system is enhanced by further processing its intermediate data, with the goal of producing a digital 3D model with a high quality color texture and an improved representation of the high-frequency shape detail. The proposed solution starts from the usual output of the scanner, a 3D model and a video sequence gathered by the scanner sensor, for which the rigid motion is known at each frame. The produced color texture is deprived of the typical artifacts that generally appear while creating textures from several pictures: ghosting, shadows and specular highlights. In the case of objects made of diffuse materials, the system is also able to compute a normal map, thus improving the geometry acquired by the scanner. Results demonstrate that our texturing procedure is quite fast (a few minutes to process more than a thousand images). Moreover, the method is highly automatic, since only a few intuitive parameters must be tuned by the user, and all required computations are particularly suited to GPU programming, making the method convenient and scalable to graphics hardware.