It should be noted that the historic character and spatial limitations of Death Cell 601 excluded the execution of the research in fully optimised conditions. Because of the delicate state of the cell and its heritage status, the conservator-in-charge strictly limited the access of people and equipment as well as research time for data acquisition. Therefore, the present study does not aim at pushing the analytical limits of non-destructive imaging, but rather on revealing the historic inscriptions.
Over the past two decades, thermography has found many applications in the study of layered objects in cultural heritage (5)(6)(7). More recent papers have shown how advanced forms of thermography have been very successful in optimizing the revelation of hidden features on (re)plastered wall paintings. Davin et al. have demonstrated the effectiveness of cooling-down infrared thermography (8). Their method showed a significant step forward in depth resolution as well as the potential to study the nature of pigments in hidden layers. Daffara et al. have combined thermography and thermal quasi-reflectography in effectively mapping both surface materials and subsurface detachments in mural paintings (9). A further integration of thermography in a multi-modal, surface and subsurface Non-Destructive Testing (NDT) campaign was shown by Sfarra et al. in the characterization of a fresco in the Church of Santa Maria di Collemaggio, heavily damaged during the 2009 L'Aquila earthquake (10). In addition to thermography, TeraHertz spectroscopy has become available, both for the subsurface study of easel paintings (11)(12), as well as plastered structures such as mural paintings (13). While some of the approaches described above would have been promising to apply to the current case study, the limitations described above -as well as the unavailability of more advanced techniques- prevented us from doing so.
Raking light photography
Raking light photography is a technique in which a specimen is illuminated at an oblique angle in relation to its surface. By doing this, the light will accentuate the irregularities located at the surface, visualizing the topography of a surface in a qualitative manner. A more quantitative approach is Reflectance Transformation Imaging (RTI). This usually involves the lighting of a surface from multiple angles with a handheld lighting source, after having placed reflective spheres near the surface to record the direction of the light. This results in an interactive display of the surface with variable lighting conditions for later study. First proposed by Malzbender two decades ago (14), the method is now routinely used in the characterization of cultural heritage such as wall graffiti (15). In our case, some inscribed characters were already visible to the naked eye and we anticipated that a small accentuation of the topography would result in legible images. That is why we chose to use raking light photography instead.
In the present study, the wall section of interest in cell 601 has been divided into 6 columns and 4 rows (24 sections). Pictures were made of every section using a Canon 500D camera with the light directed from the top, bottom, left and right side, almost perpendicular to the wall. The camera was placed on a monopod positioning it at the same height as each studied section. Due to the strict constraints of the historic cell, including the uneven floor, accurate quantification of the angles and distances was not possible. By merging the four pictures from top, bottom, left and right using Adobe Photoshop CC 2019©, a complete mapping of the surface is made. In an earlier preliminary study a part of the plastered wall section was tested to probe the feasibility of a more extensive study (16).
Thermography
Thermography is a technique in which a specimen is heated, for example by visible light, and the temperature distribution at the surface of the specimen is recorded in a thermal image, using specialised scanning cameras. If a surface with a uniform colour and texture is thus viewed with an infrared camera it will appear uniform in case the wall material is free of defects. Cracks, delamination and other defects within the wall material decrease its thermal conductivity. Consequently, at these locations the surface will heat up faster under irradiation, since the dissipation of heat is hampered. Thus, the location and morphology of the defects will become apparent in the thermal record (17). In the present study the heating is effectuated using a halogen lamp of 500 Watt. The glass lid of the lamp is removed in order to increase the intensity of infrared radiation illuminating the defected surface. Although increasing the intensity is not beneficial in thermography studies aiming to reveal subsurface defects, it does improve the detection limit for surface defects by increasing the contrast between defected and uniform areas (18). It is further chosen to move the halogen lamp manually during recording to vary the angle of incidence of the infrared radiation. The varying angle of incidence also enhances the contrast between defected and uniform areas as the radiation is reflected in different directions, depending on the topography of the surface. The detection is done with a FLIR SC7000 camera at a recording rate of ten frames per second for a maximum time of 100 seconds.
Image processing of thermography data
Due to the circumstances of acquisition for the thermography data, we could not record the configuration of the heat source over time, e.g., incident angles, location, and orientation. This obstructed direct modelling of the geometry of the underlying surface. Nonetheless, the variation of intensity observed in the thermography data provides some clues to the depth of the scratches in the surface. When processed, albeit suboptimally due to missing information, these depth clues improve readability and are thus worth investigating.
Image processing was used to distil a readable, single image of the surface from the video of thermography data. The data exhibits two informative properties: 1) defected and uniform areas respond differently to the varying angle, providing information about the topography of the surface; and 2) direct correspondences between pixels in each frame, since the sensor remained stationary. At the same time, each timestep exposes only a subset of the entire surface with sufficient visibility and the images are noisy.
An averaging operation in combination with gradient-based image editing was used to capitalise on and compensate for these properties. The underlying assumption is that the intensity values at defected areas are more stable than those at uniform areas, which are exposed by the average. Moreover, the averaging operation increases the signal-to-noise ratio (19)(20). The full process is as follows: a number of frames (2 to 4) were manually selected based on their readability. Next, a rectangular region of interest was annotated on each frame. For each selected frame, the k most similar frames were sampled, based on Euclidean distance in image space and the gradients of the k similar frames were averaged. Only the k most similar images are averaged to avoid oversmoothing and to reduce the influence from outliers. Finally, the average gradients in the region of interest for each selected frame were merged into one image using Poisson integration (21).
Lab-scale physical reconstruction of inscriptions according to sample stratigraphy
In order to test the selected Non-Destructive Testing techniques, described earlier in this section, before applying them to the plastered wall section in cell 601, a lab-scale physical model wall with inscriptions was made. For the design of this wall, several core samples from neighbouring cells in the Scheveningen prison were extracted in order to determine the stratigraphy of the cell walls and the composition of their materials. It was not possible to extract a core sample from cell 601 because of its monumental status; it is therefore assumed that the wall material of cell 601 is similar to that in its neighbouring cells.
Figure 2: a) Picture of an epoxy-impregnated and polished core sample extracted from cell 600. b), c) Schematic representations of cross-sections of the post-war plastered wall section in cell 601 and the physical model wall. The section between the dotted red line to the thicker red line approximately denotes the layer containing inscriptions.
After extraction, samples of the wall cores were impregnated in epoxy, cut vertically and polished. Figure 2a displays a picture taken with optical microscopy of the epoxy impregnated and polished core sample extracted from cell 600, the cell north of cell 601. In the eight layers visible in figure 2a, the layer of yellow paint (layer 2) is identified as the surface layer during the war. Later, paint layers 3 to 6 have been applied. The layer indicated as 5 was actually void before being infused with epoxy during sample preparation. Layers 7 and 8 are plaster layers. Figure 2b schematically shows the expected stratigraphy of the plastered section of the northern wall of cell 601: two layers that are identical or similar to the layers 1 and 2 of cell 600 plus the additional post-war plaster layer.
The phase constituents and chemical composition of the mortar of layer 1 of cell 600 were determined using X-ray diffraction and X-ray fluorescence. The mixing ratios for the mortar with which the physical model wall is made are based on these results, although slight differences were inevitable. The dimensions of the physical model wall are 0.6 m × 0.7 m × 0.02 m and it is divided into 8 sections. In each section 5 lines (1 to 3 mm deep), 8 pits (1 to 3.5 mm deep) and some letters and/or numbers (0.1 to 1 mm deep) are engraved. Two of these sections are shown in the figures 3a and 3b. On top of the mortar layer, a layer of plaster is applied (see figure 2c). The thickness of the plaster layer for the two sections shown in figures 3a and 3b varies between 0.3 mm and 0.6 mm. This layer is sanded to obtain a flat surface.
The stratigraphy of the physical model wall and the expected stratigraphy of the plastered wall section in cell 601 are very similar. The main difference is the thickness of the mortar layer, which is 2 cm for the physical model wall and approximately 1 cm for the cell wall, with approximately 10 cm of brick behind the mortar.
It can however be assumed that the thermal penetration depth in the present application is less than 1 cm. Hence, both walls can be considered infinitely thick. The mortar used in the physical model wall is similar in composition to the mortar in cell 600 of the Oranjehotel, but more finely grained. The composition of the post-war plaster in cell 601 could not be analysed and it was therefore not possible to mimic this plaster in the model wall.