This paper presents a new method to perform an illumination-based segmentation of structured surfaces. For this purpose, an illumination series is recorded, i.e. a series of images taken with directional lighting from different directions. For each location of the surface a signal is extracted from the series that describes the intensity of the corresponding point depending on the illumination angle. From this signal features are extracted that enable a pointwise segmentation of the surface. Aided by suitable models, the reflection properties of the surface can be described, and the corresponding model parameters can be estimated from the intensity signal. The influence of sampling the illumination space on the segmentation is investigated. It is desirable to keep the number of illumination angles as small as possible. On the other side, a minimum number of images is needed to achieve a reliable segmentation. The performance of the proposed approach is demonstrated with real images. The results prove that the higher data acquisition expense is compensated by a significant increase of both the spatial resolution and the robustness.
Print ISSN: 0171-8096
Volume: 73, 04/2006
Pages: 200 - 207