Details of the methodology of the archival research have been reported [32]. The difficult accessibility of the mural raised several technical challenges in capturing images capable of some iconographic elements. Portable HSI in the 400-1000nm range and Near-InfraRed (NIR) Photographic Reflectography were combined to inspect the painting in remote sensing mode, to cover the Visible and the Near Infrared spectral ranges at adequate spectral and spatial resolutions. Measurements were performed in August 2022. The room being a site of historical interest, cumbersome scaffolding was considered incompatible. A system of tripods was used to mount the different equipment (cameras and lights) in order to acquire different aspects at variable distances and orientations from the painting surface. The same illumination sources served both imaging techniques, thus reducing the in situ gear needed.
Using tripods meant sub-optimal positioning of camera and lights with respect to the target surface. This can result in both non-uniform illumination of the target area and geometrical deformation of the acquired image. These problems can affect the quality of final data; nevertheless they do not impaired the possibility of singling out details of the painting that are not appreciable by visual inspection.
a) Lighting system
A single reflector mod. ARRILTE 750Plus with a tungsten halogen lamp was placed on a tripod at a distance of about 8 metres from the surface of interest (Fig. 1). This configuration guaranteed uniform washing lighting on the target surface.
b) Near-InfraRed Photography
Infrared Reflectography (IRR) includes a class of techniques generally used to detect underdrawings [33, 34] with several options involving variable actions that depended on the sensors, optics and experimental setups [35–38]. As a simplified version of IRR, Near-InfraRed (NIR) Photography uses normal digital cameras modified to exploit the 750-1100nm region; the tail of the sensitivity range of commercial CCD Silicon detectors. Despite limited coverage, NIR Photography is effective for the preliminary inspection of the layers beneath the painted surfaces [39, 40], which was selected here because it is lightweight and highly versatile. The measurements were performed using a digital camera mod. Canon EOS RP, full-spectrum converted [41] and equipped with Canon 50mm f/1.8 STM objective lens, and full frame 26.2 Megapixel CMOS sensor, 35,9 x 24 mm size. The modification consisted in the removal and replacement of the native camera’s low-pass filter with an IR pass filter at 850nm. The matrix sensor guaranteed excellent image quality in terms of spatial resolution, and therefore image definition, even under remote shooting conditions. The tripod camera was positioned at different distances (3.5 metres down to 50 cm) to take images with different spatial resolutions.
c) Portable VNIR Hyperspectral Imaging
HSI uses a sequence of spectrally contiguous reflectographic images acquired throughout the sensitivity interval of the HSI camera. The output is a data-set (image cube), in which each pixel (point) of the framed image is associated with a reflectance spectrum [42]. It is ideal for the non-invasive investigation of polychrome surfaces providing compositional information on pigments and artists’ materials, while allowing production of detailed images like distribution maps and false colour images which highlight several features not evident to the naked eye. The acquisition enables reconstruction of colour RGB images along with inspection of inner features in the NIR. Several pigments are also identifiable based on their Vis-NIR spectra. Since the measurements are performed in reflectance mode, proper illumination of the target is required, typically a broad band lighting source emitting in the entire Vis-NIR-SWIR range. The intensity of incident radiation is also crucial to guarantee proper signal to noise ratio. The acquired HSI data are big size sets embedding redundant information. Their analysis is based on suites of algorithms for data-reduction and multivariate analysis, selected on the basis of the specific applications. Among the most used algorithms for cultural heritage applications are principal component analysis (PCA), minimum noise fraction (MNF) and the classification method Spectral Angle Mapping (SAM). These methods generate as outputs set of new images, which perfectly overlap with the RGB colour image, and are expressed as false colours pictures and maps which highlight paint layer inhomogeneities on the surface or elements that are often not evident on visual inspection.
A portable HSI camera push broom type mod Specim IQ was used. The camera is compact (207 x 91 x 74 mm) and light (1,3 Kg) mounted on a tripod, and operated at variable distances from 10 cm to tens of metres. The nominal spectral range is 400-1000nm, with 7 nm spectral resolution. The sensor is a 512 x 512 pixel. The F/number is 1.7. The optimal distance was established on the basis of operational needs, considering that as the distance increases, the framed area increases but reductes the spatial resolution, and vice versa. Various shots were taken at distances of a few metres from the surface orientated to frame the areas of interest entirely but to the detriment of the spatial resolution. This configuration resulted in poor detail quality of HSI images but nevertheless highlighted the distribution maps of the materials. The poor spatial resolution was compensated for by the higher spatial resolution of NIR photography. The white reference calibration is crucial in the HSI remote-sensing acquisition mode. The measurements were performed by imaging a certified 99% Reflectance Spectralon target, which was included within the acquisition field without contact with the fresco. A tripod with high elevation was used.
d) Data processing
Data processing of HSI data-cubes acquired with Specim IQ was performed using the ENVI © software [43].
Any possible material inhomogeneities or hidden feature was investigated using PCA and MNF methods.
The PCA and MNF algorithms were applied to pre-treated HSI data-cubes, which were spectrally cut at the edges of the range to exclude the noisy contributions. Analysis was then performed in the 420-900nm range.
The IR data images acquired with the Canon camera were processed using XnView MP free software [44].