Hyperspectral imaging and spectral classification for pigment identification and mapping in paintings by El Greco and his workshop

Abstract

The identification of painting materials is of essential importance for artistic and scientific analysis of objects of artistic and historic value. In this paper we report a new method and technology comprising a) hyperspectral imaging, b) development of spectral libraries corresponding to target materials and c) proper classification strategies with (a) and (b) as inputs. Our findings advocate that the method improves radically the diagnostic potential of visible-near infrared imaging spectroscopy. A system’s approach is implemented by combining a novel hyperspectral camera integrating an innovative electro-optic tunable filter solution with spectral analysis and classification algorithms. A series of pigment material replicas was developed using original methods covering almost the entire palette of Renaissance painters. Hyperspectral acquisition of the constructed pigment panels provided millions of spectra, which were used for both training and validation of a series of spectral classification algorithms, namely: Maximum Likelihood (ML), Spectral Angle Mapper (SAM), Normalized Euclidean Distance (NEUC), Spectral Information Divergence (SID), Spectral Correlation Mapper (SCM) and Spectral Gradient Mapper (SGM). It was found that the best performing algorithm in identifying and differentiating pigments with similar hue but different chemical composition was the ML algorithm. This algorithm displayed accuracies within the range 80.3%–99.7% in identifying and mapping materials used by El Greco and his workshop. The high accuracy achieved in identifying pigments strongly suggest that the new method and technology has great potential for the scientific analysis of artwork and for assisting conservation and authentication tasks.