Artefact Matching and retrieval Using the Generalised Hough Transform

One of the main tasks of the archaeologist is concerned with the classification or identification of artefacts and much previous work has been reported on the application of computer technology to the storage and retrieval of artefact shape data. The bulk of this work has been based on artefact profiles following the manual approach to artefact classification using line drawings of outlines. For example, Hall and Laflin (1985) used the B-splines curve-fitting technique to generate object ouüines, and suggested that this method could be used for storing and matching profiles. Gero and Mazzullo (1984) used closed-form Fourier series to describe an object's shape, and suggested that the Fourier series could be used for matching. Peter Main has carried out much work on artefact shape: on databases for archaeological artefacts, representation of their shapes by tangential profiles schemes, and on artefact retrieval by shape from such databases (Main 1986, 1987). In this paper we describe a prototype System for Matching ARTefacts, (SMART). It uses a versatile approach to artefact classification based initially on the generalised Hough Transform (GUT) although a range of classification techniques are anticipated, including texture which is discussed briefly in section 4.6.4. The possible use of the GHT for artefact shape matching was proposed in a paper by Lewis and Goodson in 1990 (Lewis & Goodson 1991) as it is a robust matching algorithm capable of producing a response when only a partial shape is available. Here we demonstrate the effectiveness of the technique not just as a tool for profile matching using line drawings, but for a range of artefact retrieval and matching tasks. These include retrieval of similar shapes when only a fragment is available and retrieval of line drawings when the input image is from a photograph. One of the other benefits of the approach described here is that the shape data for the database does not have to be extracted and represented explicitly either by manual digitisation of the outlines or by some other means. A raster image of the drawing or the photograph is simply captured into the image database either by scanning the paper line drawings or the photograph using a flatbed scanner or by using a video camera and digitiser. No further pre-processing is needed before the application of the algorithms. Section two of the paper describes the way in which the SMART system appears to the archaeologist, in other words the current user interface for the search system and the options available. The third section describes the methods used for handling the images. Section four describes the details of the generalised Hough Transform and the techniques for calculating the similarity index used for establishing the matches. The fifth section provides data from a test set of images and shows the results achieved and in the final section proposed enhancements to the SMART system are discussed.