Application of Computer Vision algorithms for automatic classification of archaeological artefacts

Edgar Francisco Román-Rangel, Diego Jiménez-Badillo


The application of computer vision technologies for the analysis of cultural heritage artefacts has witnessed a rapid growth during the last decade. This is especially true with regard to the creation and use of digital 3D models, which enable capabilities that would not be available using the original artefacts, such as automatic and semi-automatic content analysis, virtual reconstructions, more efficient archiving, sharing documentation online, training of novel scholars, etc.
An area of especial interest is the statistical analysis of shape features observed on 3D models of artefacts, especially ceramic vessels and pottery sherds, with the purpose of categorizing and classifying objects in an automatic way.
In this paper we present new results of an on-going project focused on applying computer vision techniques for automatic classification of archaeological artefacts. We discuss some useful approaches that involve the extraction of shape descriptors (SIFT, Spin Images, etc.) within a Bag of Visual Words model and propose a novel technique for local description of 3D surfaces called Histogram of Spherical Orientations (HoSO).
The HoSO local descriptor consists of the quantization of the local orientations of a point with respect to its nearest neighbours. Such local orientations are computed both in the azimuth and the zenith axes. The frequencies of the local orientations are stored in a histogram, which can then be used for comparison and matching purposes