Relief Extraction from Stone Monuments

Abstract

In this paper, we propose a method to extract reliefs inscribed on stone monuments by analyzing 3-D scan data. Since a relief (carving) is canal-shaped, the curvature-based Frangi filter that has been used in blood vessel detection is suitable for relief extraction. The curvature-based Frangi filter evaluates how well a local surface forms a canal-like curve using principal curvatures. However, for rough surfaces on old and weathered stone monuments, estimated curvatures tend to be noisy, and thus, relief extraction results are degraded. In the proposed relief extraction method, after an initial relief candidate search using curvature-based vesselness estimation, we evaluate the reliability of each relief candidate. Then, for each relief candidate, its vesselness is weighted by the corresponding reliability. As a result, false positives caused by dents and scratches in the relief candidates are successfully alleviated. We confirm that the proposed method extracts reliefs clearly (even from uneven, rough surfaces) and outperforms the conventional curvature-based Frangi filtering method.