CAA2016 has ended
Thursday, March 31 • 08:55 - 09:20
S06-02 A machine learning approach for 3D shape analysis and recognition of archaeological objects

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

A machine learning approach for 3D shape analysis and recognition of archaeological objects

Diego Jiménez-Badillo, Mario Canul Ku, Salvador Ruíz-Correa, Rogelio Hasimoto-Beltrán

Museum professionals all over the world have always shown great interest in acquiring automatic methods to register and analyse the shape of cultural heritage artefacts. Thanks to recent advances in 3D scanning and photogrammetry techniques, it is now possible to model the surface of objects with very little effort and in a relatively short time. 
The continuous adoption of these techniques in cultural institutions has generated thousands if not millions of 3D digital models. Unfortunately, after these resources are produced, very little effort is spent in making them accessible to researchers or the general public. Part of the problem is a lack of efficient computer mechanisms to search, retrieve and classify 3D data.
The conventional way to search and retrieve 3D models consists in composing a query based on text descriptions. However, textual annotations are necessarily constrained by the database application domain, ontology, etc., as well as by language and other factors. Consequently they are inadequate for shape oriented searches. 
This paper presents results of an on-going project focused on developing a computer platform to automatize the search, retrieval, recognition and analysis of 3D object models. The platform processes queries based on geometric properties instead of text. Simply stated, the computer program takes a 3D surface mesh as input (i.e. the query model). Then, a search engine compares it to hundreds or even thousands of 3D scanned objects stored in a repository identifying those that approximate the shape of the query model. Next, the matching models are retrieved, ranked by degree of similarity and displayed to the final user. Afterwards, additional tools can be deployed to perform some kind of analysis on the objects retrieved. 
A platform like this is much more powerful than a text search engine because it avoids mismatching situations, such as when a person queries the database looking up for “bowls” and retrieves nothing just because the bowls are labelled as “cuencos” (a Spanish term) or “cajetes” (i.e. a term common in Mesoamerican archaeology to described the same type of vessels). Moreover, the platform is able to exploit mathematical analysis algorithms for automatic classification of shapes.
During the presentation, we discuss the specific requirements that a shape recognition platform must satisfy to be useful in museums and cultural heritage research. In archaeological projects, for example, we encounter objects that are not necessarily identical in terms of geometry and yet they are considered to belong to the same class. We also intent to show the first part of this platform, namely the search engine for matching and retrieval of 3D Objects.

Thursday March 31, 2016 08:55 - 09:20 CEST
Domus Academica, Theologisk eksamenssal