Machine Vision and Applications (2010) 21:989–998 DOI 10.1007/s00138-009-0198-7 SHORT PAPER A novel framework for 3D reconstruction and analysis of ancient inscriptions Angelos Barmpoutis · Eleni Bozia · Robert S. Wagman Received: 16 January 2008 / Revised: 30 January 2009 / Accepted: 13 April 2009 / Published online: 26 May 2009 © Springer-Verlag 2009 Abstract Studying ancient inscriptions is based up to date mostly on observation and manual analysis by means of which epigraphists attempt to establish a geographical and chronological classification as well as to analyze the lettering techniques. In this paper we propose a novel framework for efficient 3D reconstruction of inscriptions and for statistical analysis of their reconstructed surfaces. The proposed frame- work employs a shape-from-shading technique to reconstruct in 3D the shape of the inscribed surfaces. The obtained sur- faces are segmented into smaller box-shaped regions contain- ing single letters. These letters are classified into groups of same characters or symbols and then an atlas (average) letter shape is created for each character. For the construction of those atlases we employ a functional minimization method that registers the surfaces of same letters to the unknown aver- age surface, which is also estimated simultaneously. Using the estimated atlases, an automated analysis of the inscribed letters is performed. This framework can be effectively used for the study of the variations of the lettering techniques within an inscription or a set of inscriptions. We applied our framework to five ancient Greek inscriptions. Our results are reported in detail and the variations found in lettering tech- This research was in part funded by the Gerondelis Foundation Inc. grant to EB and the University of Florida Alumni Fellowship to AB. A. Barmpoutis (B ) Computer and Information Science and Engineering, University of Florida, Gainesville, FL 32611, USA e-mail: abarmpou@cise.ufl.edu E. Bozia · R. S. Wagman Department of Classics, University of Florida, Gainesville, FL 32611, USA e-mail: bozia@ufl.edu R. S. Wagman e-mail: rsw@classics.ufl.edu niques are commented on by archaeologists who also validate the accuracy of our proposed method. Keywords Epigraphy · Squeeze digitization · Shape from shading · Atlas construction · Agglomerative hierarchical clustering 1 Introduction The evolution of imaging technology has affected positively several areas in the computer vision community. During the last decades the significant progress in sensoring technol- ogy has helped us capture high-resolution multimodal images that have consequently resulted in the advancement of med- ical imaging (magnetic resonance images [2, 23]), general image and video processing [8, 22, 26], and virtual 3D envi- ronment synthesis (for instance, reconstruction of archaeo- logical monuments [7, 18, 19, 21]). In archaeology a crucial field of research is the study of ancient inscriptions. Some questions that archaeologists are called to answer pertain to the dating of the inscriptions, the attempt to identify the place of origin, and the analysis of the lettering techniques. So far the method that has been com- monly used for the study of inscribed fragments is the fol- lowing: the archaeologists use a special type of moisturized paper (squeeze) which they push on the inscribed surface using a brush specially adapted for the purpose. When the letters are shaped on the squeezed paper, the archaeologists let it dry, creating that way an impression of the inscription (see Fig. 1). The paper most in use is a type of filter paper obtained from laboratory suppliers and it may vary in thick- ness. A thinner paper needs less working on the stone, but may disintegrate more easily [27]. Other media have been also used in literature for this purpose, for instance latex, 123