Supervised segmentation of 3D cultural heritage

The use of 3D models for the documentation and dissemination of cultural and archaeological heritage is widespread today. Nevertheless, to provide useful 3D data, it is important to associate semantic information that can help operators understand the heritage. This study aims at carrying out an optimal, repeatable and reliable segmentation procedure to manage various types of 3D survey data and associate them with heterogeneous information and attributes to characterize and describe a surveyed object. The developed method starts from 2D supervised machine learning segmentation of orthoimages or UV maps and then projects the segmentation results on the 3D data. Three case studies are presented to demonstrate that the proposed approach is effective and with further potential e.g. for restoration and documentation purposes.