3D Pottery content-based retrieval based on pose normalisation and segmentation

This paper presents a novel compact shape descriptor designed specifically for content-based retrieval of complete or nearly complete 3D vessel replicas. The descriptor consists of two vectors that carry morphological features of the vessel's main body and appendages. The extraction of the descriptor is based on a pose normalisation preprocessing phase which is designed for axially symmetric objects. In order to evaluate the efficiency of the descriptor, we created a calibrated ground-truth database of 1012 3D digitised and manually modelled vessels and performed multiple query-by-example experiments. We present the performance of our descriptor in relation to the performance of the MPEG-7 3D shape spectrum descriptor. Additionally, a web-based 3D content-based retrieval prototype system has been developed based on open source technologies.