A hyperspectral imaging spectral unmixing and classification approach to pigment mapping in the Gough & Selden Maps

Abstract

ABSTRACT In this research, a spectral unmixing and classification approach for hyperspectral imagery (HSI) of historical artifacts is introduced. The Gram Matrix and Max-D techniques have been used to estimate the within material diversity in an HSI and extract the spectra of the class exemplars, called endmembers. Due to the absence of ground truth, the exemplar spectra are assumed to represent pure materials in the scene or unique pigment mixtures. After extracting endmembers, the Nonnegative Linear Least Squares algorithm and the K-means algorithm are then used to classify the HSI based on the endmembers. We also proposed a “spectral angle spatial patterns” method to map the class membership and goodness across the map to identify per-class spatial patterns. This approach has been successfully utilized in two historical artifacts, the Gough Map of Britain (c.1400) and the Selden Map of China (c.1619). Both maps were imaged using a hyperspectral imaging system while in the collection at the Bodleian Library, Oxford University. It reveals at least six kinds of dominant water pigments used in each map along with their spatial distribution. This approach can be generalized to a novel pigment mapping tool for historical geographers to analyze the material diversity of historical artifacts.