Virtual reconstruction of archaeological vessels using convex hulls of surface markings

This paper presents a 3D computer vision method that assists the tedious procedure of manually reconstructing ceramic vessels from fragments unearthed in an archaeological excavation. This computational method relies on vessel surface markings combined with expert feedback (via the archaeologist) to form a generic model of a vessel that the excavated fragments might have originated from. Prior expert information (i.e., the generic vessel model) is usually based on a host of factors ranging from historical knowledge from the period in question to the provenience of the excavation, etc. The generic model need not be exactly identical to the excavated vessel, but rather within the allowable class (i.e., within a geometric transformation of it in most of its parts). The ceramic vessels we deal with here have surface markings. We exploit the 3D convex hull associated with the surface markings as they are preserved under the allowable set of transformation (between the generic expert model and the excavated vessel). We align these using a novel set of discrete cross weighted moments. The morphing transformation (affine or higher order morphing function) is computed from these corresponding convex hulls and their corresponding vertices, and distance error metrics are introduced to access the fitness of the alignment of a fragment to a given vessel. The methods are tested on a subset of ceramics excavated from Independence National Historical Park (INHP) created through 3D scanning of prospective generic bowls and their pieces.