A Study of Spectral Imaging Acquisition and Processing for Cultural Heritage

Sony George, Jon Y. Hardeberg, João Linhares, Lindsay MacDonald, Cristina Montagner, Sérgio Nascimento, Marcello Picollo, Ruven Pillay, Tatiana Vitorino and E. Keats Webb


Imaging spectroscopy (IS), specifically multispectral (MSI) and hyperspectral imaging (HSI) techniques have proven to be effective  non-invasive analytical tools for material cultural heritage. The combination of digital imaging and spectroscopy expands point-based, or one-dimensional, spectroscopic techniques. IS results in the ability to map the spatial distribution of materials over an entire object, extract reflectance spectra for the identification of materials, enhance and reveal underdrawings, identify past conservation treatments, and measure colour. Increased application of these techniques to conservation documentation calls for the standardization of methodologies and definitions of best practices, to allow institutions to have reproducible and comparable data. Moreover, there is a need for a better understanding of the instrumentation, the elements of data acquisition, and the accuracy and reliability of the data for different instruments and methodologies.

The development and increased application of MSI and HSI to documentation of cultural heritage is acknowledged and addressed through the European Cooperation in Science and Technology (COST) Action, Colour and Space in Cultural Heritage (COSCH) and the tasks of its Working Group 1 (WG1). Focusing on spectral object documentation, WG1 had the task of identifying, characterizing and testing spectral imaging techniques and devices in the visible and near infrared range. To assess the various spectral imaging systems and working towards standardized methodologies and best practices for cultural heritage, WG1 carried out a Round Robin Test (RRT). Five objects were recorded by nineteen institutions with various MSI and HSI systems and setups. The objects included a SphereOptics Zenith Polymer Wavelength Standard, an X-Rite ColorChecker Classic, a painted panel reconstructed using traditional fifteenth-century Tuscan techniques, an antique Russian icon on a tinned steel plate, and a white card. The RRT was a coordinated research effort aiming to gain a better understanding of the instrumentation, the elements of data acquisition, and the effects of the instruments and methodology on the accuracy and reliability of the data. In addition, important aspects of the scientific reliability and accuracy of spectral image data are reproducibility and comparability. Reproducibility emphasises the ability of an institution and device to produce the same results under similar conditions, while comparability emphasizes the ability to compare data over time (as part of monitoring) or between different institutions.

The work of summarizing and visualizing the received RRT data has exposed the challenges and complexity of the assessment and comparison of the different data sets. Understanding the variation in the resulting data sets helps with defining best practices for Cultural Heritage. The experience was a means of working towards optimised methodologies in the application of non-contact, high-resolution techniques to the state-of-the-art documentation of cultural heritage.

Keywords: Imaging spectroscopy, polychrome surfaces, reflectance, spectral image quality, image colour accuracy, cultural heritage imaging, calibration workflow, multispectral, hyperspectral, COSCH



Figure 1: a) X-Rite ColorChecker with sampling areas for colorimetric and spectral analysis. b) Russian icon with sampling areas for colorimetric and spectral analysis. c) SphereOptics Zenith Polymer Wavelength Standard. d) Replica panel painting with sampling areas for colorimetric and spectral analysis.



Figure 2: Different spectral imaging devices used within RRT. Left MSI; right HSI.



Figure 3: Minolta CM-2600d measuring reflectance spectra on selected areas of the Russian icon.



Figure 4: ColorChecker, comparison between the four imaging devices (MSI= 1, 2; HIS= 3, 4) and the spectro-colorimeter measurements: (a) DE00 and DC’, (b) RMSE, and (c) GFC data.



Figure 5: DeltaE00 for the 24 colored areas of the ColorChecker. Average results for MSI and HSI devices.



Figure 6: Russian icon, comparison between the four imaging devices (MSI= 1, 2; HIS= 3, 4) and the spectro-colorimeter measurements: (a) DE00 and DC’, (b) RMSE, and (c) GFC data.


Figure 7: Russian Icon detail extracted from the different devices to show the spatial resolution quality from three IS systems.



Figure 8: Reflectance spectra from two different colored regions (skin and blue vest, respectively) of the Russian icon reconstructed from the MSI and HSI datasets.



Figure 9: Rare-Earth Wavelength Standard. Comparison of Vis-NIR (left) and SWIR (right) Reflectance.



Figure 10: Comparison of the reflectance spectra of wavelength standard for one of the Vis-NIR hyperspectral systems (top left), a zoom of a narrow range of wavelengths showing spectral misalignment (top right), the derivative of the spectra (bottom left) and the residual error with its linear regression (bottom right). Reference = grey line; Vis-NIR hyperspectral system = black line.



Figure 11: Detail of the replica panel painting extracted at approximately 1040 nm looking at the line patterns to visually assess the system’s ability to resolve fine details.



Figure 12: The vertical profile (left) from a HSI device data looking at the exposed watercolor lines (right image) on the test panel. The system is able to differentiate between the lines and spaces as seen with the well-defined peaks.



Figure 13. Images extracted from the hyperspectral data-cube at two different wavelengths for the visualization of underdrawing details made with lead- and tin-based metalpoint (left) and watercolor (right) techniques.



Figure 14: Reflectance spectra from the painted areas made with carmine lake (dotted line), azurite (dashed line), lead white (solid line), and gypsum preparation (grey line) of the replica panel painting.