Improving high-speed scanning systems by photometric stereo

High-speed scanning systems can be extremely valuable for Cultural Heritage applications, especially when large collections of small objects have to be acquired. However, fine details may not be acquired using this technology. Nevertheless, it is possible to try to recover them by taking advantage of the additional data provided by these systems: the calibrated video sequence of the acquisition, and the position of the projector light for each frame. In this paper, we propose a workflow that processes the video sequence with a photometric stereo approach, in order to refine the coarse geometry provided by the scanner. A normal map is first extracted by a method that accounts for the unevenly distributed sampling that generally results from the particular trajectory followed by this kind of scanners during the acquisition. This normal map is then integrated in order to recover missing geometric features. Good performances are achieved, since the whole workflow is particularly suited to GPU programming.