No document available.
Abstract :
[en] Over the past few years, the spectral classification gained major interest in the field of teledetection, urban landscape and some cultural heritage application in order to detect non straight visible features. On the other hand, 3D acquisition lead to the gathering of high resolution point clouds describing the geometric object’s features. Both approaches are rarely combined in the archaeological domain despite the fact that an increasing number of museums are digitizing their collection in 3D and perform two dimension multispectral analysis in conservation, analysis and restoration purposes. In this study, we propose to attach multispectral information to a point cloud representing archaeological objects. A multivalued classification can then be performed to the cloud. The approach shows two major interest: first, the three spatial dimensions of the point cloud can be considered in order to improve the classification; secondly, since the classification results are known in 3D, the classes can be easily represented in 3D which is a major issue for archaeological results interpretation. The artefacts digitization goes through a photogrammetric process. We produced objects images in multiple wavelengths ranging from ultraviolet, visible to near-infrared spectrum. From the 3D textured mesh (one per wavelength), we applied a data fusion to generate a unique point cloud containing the multispectral information as scalar fields. We tested several unsupervised and supervised classification processes dedicated to multispectral information or not of the point clouds such as K-means, Random Forests… The classification results allows to segment the point cloud and highlight the discovered parts of the artefact. Based on the training classification set, researchers can enhance the different features of an object (pigments, composite objects, restorations). The segmented data allow automatic surface repartition measurements and a better understanding of the creation process.
