Mixing to unmix: painting component identification in historical graphic documents via data fusion of DRIFTS and VNIR-SWIR reflectance spectra

Abstract

The identification of historical painting materials is essential for understanding artistic techniques, proposing conservation strategies, and supporting dating and provenance studies. However, the presence of pigment and dye mixtures, along with complex pigment–binder interactions, poses significant challenges for non-invasive analysis. This study presents a spectral unmixing approach based on the data fusion of hyperspectral imaging (HSI) and Diffuse Reflectance Fourier Transform Spectroscopy (DRIFTS) data to identify the painting components of mock-up samples that replicate historical objects. Results show that fusion significantly improves the identification of mixture painting components compared to single-technique analysis. The complement of the Goodness-of-Fit Coefficient (cGFC) outperformed other metrics, achieving the highest rate of correct painting component identification and lowest error for the greater presence of some elements compared to others. Preprocessing steps including Savitzky–Golay derivative, spectral cropping, and normalization proved essential for maximizing performance. The method was further validated on a set of historical manuscripts, correctly identifying painting materials in the majority of cases.