Near infrared spectroscopic reflectance imaging: supervised vs. unsupervised analysis using an art conservation application

Near-IR spectroscopic imaging was used to analyze the remnants of a work of art, a 16th century drawing, attributed to the School of Pieter Bruegel the Elder, which had been significantly altered during a cleaning attempt. Using a combination of a CCD camera and a liquid crystal tunable filter (LCTF), near-IR spectroscopic images (650-1050 nm) were collected from the drawing and from a test sample composed of four substances with differing near-TR spectra deposited on a whiteboard surface. Both supervised and unsupervised classification methodologies (linear discriminant analysis (LDA) and fuzzy C-means (FCM) clustering, respectively) were used to analyze the data. FCM clustering, in combination with several spectral normalization routines, proved an excellent data exploration method for the test sample. LDA gave consistently clearer results than the FCM methods, but required a priori knowledge of the spectral properties of the sample, provided, in this case, by the FCM analysis. LDA of the spectroscopic image of the work of art revealed clearly and for the first time the location of regions of the drawing where faint traces of ink residue remained. (C) 1999 Elsevier Science B.V. All rights reserved.