Segmentation of Multispectral Images of Works of Art through Principal Component Analysis

Investigation of materials constituting painted layers of works of art (panels, canvas, frescoes) can be profitably done by means of non-destructive optical techniques based on the analysis of reflectance spectra in the visible and near infrared regions. Accurate and high spectral resolution measurements can be obtained by means of fiber optics spectrophotometers, but only in small spot areas. Image spectroscopy systems can give instead a complete spectral information on the whole examined surface in a great number of bands, allowing direct visual interpretation. The analysis of such amount of data is not trivial. A possible approach is to decorrelate the data and concentrate the significant information in few images, by using principal component analysis (PCA). In this work segmentation is investigated in order to partition the imaged scene into regions of spectral similarity to facilitate successive analysis. The results on a test tempera panel and on a predella painted in the XVI century show the effectiveness of the proposed approach, also revealing details undetectable by conventional techniques.