Toward a framework for assembling broken pottery vessels

Abstract

This paper addresses how to automatically reconstruct pottery vessels from a collection of sherds using a variety of features and their comparisons. To solve the problem, we designed a computational framework that is founded on the primitive operations of “match” proposal and evaluation. A match defines the geometric relationship between a pair of sherds. This framework affords a natural decomposition of the computation required by an automatic assembly process and provides a concrete basis to evaluate the utility of different features and feature comparisons for assembly. Pairwise matches are proposed and subsequently evaluated by a series of independent feature similarity modules. Assembly strategies are abstracted from the feature-specific sherd details and operate solely in terms of the probabilistic output of pair-wise proposals and evaluations. Our framework, which is modular and extensible, paves the way for a system to automatically reconstruct pottery vessels. We demonstrate a greedy assembly strategy that predicts likely pairs and triples of sherds using a handful of proposal and evaluation modules. Previous attempts to automate the task of reconstructing pottery vessels have relied on a single feature, and sometimes user intervention, to direct the search (Ucoluk & Toroslu 1999), (Papaioannou, Karabassi, & Theoharis 2001), (da Gama Leito & Stolfi 1998). While (Cooper et al. 2001) accounts for more than one feature using complex parametric models, we propose a conceptually simpler and modular system for integration akin to (Pankanti, Jain, & Tuceryan 1994) and (Keim et al. 1999; Keim, Shazeer, & Littman 1999). Our framework and assembly strategy are similar to (Jepson & Mann 1999) where they search for a plausible scene interpretation.