Multispectral and Multi-modal Imaging Data Processing for the Identification of Painting Materials

Nowadays there are several multispectral acquisition systems which have been proposed for imaging artworks. With multispectral imaging it is in principle possible to e.g. achieve higher color quality, however what is perhaps most interesting it that it paves the way to image spectroscopy, where the final goal is to achieve the spectral signature of each imaged element of the artwork. This is a crucial step to identify pigments and binding media in a totally unobtrusive way. Whatever the system used to acquire multispectral images is, the information gathered must be processed, and an appropriate algorithm must be used to differentiate the areas within the image which have similar spectral signature and therefore similar chemical composition. The more efficient the algorithm used, the more accurate the results. However, when the system chosen has a limited number of filters, still some problem of metamerism may appear, and the algorithm used may not be able to give a correct answer in all cases. In order to have a more efficient use of a system with a limited number of filters, and therefore a quite coarse approximation of the reflectance spectral profile of the materials on the artwork, we chose to include different modalities. In order to obtain a larger amount of data, we analyzed also the radiation reflected by the artwork in the IR range, and the fluorescence emitted in the visible range when the radiating source was UV light. In this way, we could profitably compare the available data and search the same profile registered at a given location all over the painting surface. Having more modalities allowed a more reliable segmentation of the image.