Flexible bump map capture from video

Abstract

We extend recent techniques for the automated capture of surface normal maps or bump maps from real-world material samples. In contrast to recent systems, which have required cumbersome laboratory setups with careful calibration, we propose a moving-camera system which requires only one light source, rigidly attached to the camera. We describe a simple but accurate image-based calibration technique which allows the construction of an approximation to the standard photometric stereo setup. We show how to optimally estimate the system geometry during capture, and demonstrate that the accuracy of the new system is sufficient to allow normals to be computed.