Soft and hard modelling methods for deconvolution of mixtures of Raman spectra for pigment analysis. A qualitative and quantitative approach

Raman spectroscopy provides a means for the detection and identification of pictorial materials on artworks. As a non-destructive, applicable in situ and non-ambiguous technique, it is one of the most preferred to analyse the pigmentation of any kind of artwork: from paintings [1] and papyrus [2] to polychromes on woods [3]. A common problem, however, is lack of spatial resolution on some systems due to large focal distances, which degrades the theoretical high resolution of the system, which involves the resolution of mixtures of individual Raman spectra. In this work, we will present the advantages and disadvantages of two methods for the separation of mixtures of Raman spectra [4] and [5], and we present a new solution to overcome the problems of the above. To such an end, we will provide qualitative (identification of the species) and quantitative (determine their concentration profiles) results of the methods. The experimental analyses have been carried out in two steps: first we calibrate the methods with known mixtures of two compounds prepared in the laboratory. Second, we test the methods with a real artwork supposed to be from ‘El Greco’. Procedures to minimise problems, such as extreme fluorescence and noise, that arise on real artworks are also presented.