Thermal Reading of Texts Buried in Historical Bookbindings

Abstract

In the manufacture of ancient books, it was quite common to insert written scraps belonging to earlier library material into bookbindings. For scholars like codicologists and paleographers, it is extremely important to have the possibility of reading the text lying on such scraps without dismantling the book. In this regard, in this paper, we report on the detection of these texts by means of infrared (IR) pulsed thermography (PT), which, in recent years, has been specifically proven to be an effective tool for the investigation of Cultural Heritage. In particular, we present a quantitative analysis based, for the first time, on PT images obtained from books of historical relevance preserved at the Biblioteca Angelica in Rome. The analysis has been carried out by means of a theoretical model for the PT signal, which makes use of two image parameters, namely, the distortion and the contrast, related to the IR readability of buried texts. As shown in this paper, the good agreement between the experimental data obtained with historical books and the theoretical analysis proved the capability of the adopted PT method to be fruitfully applied, in a real case study, to the detection of buried texts and to the quantitative characterization of the parameters affecting their thermal readability.

1. Introduction

In the field of Cultural Heritage (CH), non-destructive evaluation techniques [1,2,3] are increasingly used since they allow researchers to gather relevant information about the investigated items, such as that concerning their structures or the processes adopted in their manufacture. In this respect, it is worth noting that most of the valuable information is quite often obtained from features located beneath the sample surface that cannot be probed by means of ordinary optical inspection techniques. Consequently, in recent years, there has been considerable research work devoted to the development of experimental techniques for the investigation of features buried inside artworks, which, among other techniques, has led to the establishment of active Infrared Thermography (IRT) as a very effective tool for this kind of study [4,5,6,7]. The IRT working principle relies on heating the sample, typically through the absorption of visible (VIS) light, and on the subsequent locally resolved detection of the induced variation in the sample’s infrared (IR) emission by means of an IR camera [8,9]. As in the present study, the pulsed IRT configuration (PT), which makes use of VIS light pulses delivered by flash lamps to heat the sample, has become the preferred one for CH surveys because of its simple and fast use. In this case, the IR camera is employed to record a sequence of images, which, in the following, are referred to as thermograms, describing the map of the local thermal state of the sample at different delay times from the onset of the heating pulse. One of the peculiar abilities of PT is given by the possibility of distinguishing between features located at different depths in the sample, with such a possibility not granted by most of the other IR techniques employed in the CH field, such as IR reflectography. In this respect, it is worth noting that thermography enables the detection of subsurface features, provided that these are located within the diffusion length of the induced temperature rise. In fact, the presence of local subsurface inhomogeneities may be responsible for introducing significant modifications in the corresponding heat diffusion rate, thus leading to a variation in the IR emission from the surface area above the inhomogeneity with respect to the surrounding area and, consequently, to the appearance of thermal contrast in the recorded thermographic images. Therefore, the analysis of thermograms obtained while varying the temperature diffusion length may allow for discrimination between features located at different depths in the sample. Such variation can be obtained by either varying the modulation frequency in the case of periodic heating, such as in lock-in thermography (LIT), or allowing the heat to diffuse over longer time durations and, hence, over larger depth into the sample in the case of step heating or PT configurations. Specifically, as regards the investigations shown in the present manuscript, LIT has also been proven to be an effective tool for the detection of buried texts [10]. However, to date, a theoretical model for the quantitative analysis of experimental LIT results obtained from buried graphical features has not yet been developed. In addition, as mentioned previously, depth-resolved LIT investigations require different measurements to be carried out while varying the heating modulation frequency. In this respect, PT may be considered an extremely valid tool since it enables thermographic inspections to be executed through a faster experimental procedure with respect to LIT.

In order to exploit PT’s ability to carry out quantitative depth-resolved studies on CH items, over the last several years, research efforts have been devoted to the development of theoretical models for the analysis of the time dependence of the PT signal. According to such models, the relation between the PT signal and the induced temperature distribution in the sample depends strongly on the optical properties of the sample. In optically opaque samples, such as those made of metallic materials like bronze statues, both the VIS light absorption and IR emission take place at the sample surface, and, therefore, the PT signal is merely proportional to the temperature variation at the sample surface [5]. Consequently, the time dependence of the IR emission is solely determined by the sample’s thermal properties, with the optical ones only responsible for affecting the PT signal amplitude [5]. Conversely, in artifacts made of semi-transparent materials, such as books, both the VIS light absorption and IR emission processes occur over the sample volume, and, therefore, the sample’s optical and thermal properties both play a crucial role in establishing the time variation in the PT signal [10]. Consequently, the modeling of the PT signal is less straightforward than in the former case since it arises from the interplay between optical propagation and thermal diffusion processes [11,12].

The present manuscript concerns a particular category of CH, that of book heritage, mostly made of semi-transparent materials. In the literature, many studies, based on the use of numerous techniques, have been dedicated to the characterization of library materials and, in particular, to the recovery of lost writings, such as erased and overwritten texts in palimpsests [13]. In these applications, ultraviolet illumination [14], hyperspectral imaging [15] and multispectral imaging [16,17] have been used for the recovery of damaged or censored texts, for the study of burned papyri and parchments and for the reading of erased inks on parchment [14], respectively. Of particular importance among these studies are those concerning the reading of texts buried in the bookbindings of historical manuscripts [10]. In fact, from the 16th century onwards, it became very common to use earlier book materials for the production of new bookbindings due to the increase in book production caused by the invention of printing [18]. In particular, reused written fragments were applied to support the structure between the board and the spine or inserted between the cover and the endpapers to strengthen the connection between the binding and the book block [19]. However, these interesting fragments are often not accessible by a simple visual inspection, because they are mostly hidden between the book cover and endpapers. It is therefore of utmost importance to have a technique that enables the characterization of these fragments without dismantling its bookbinding, revealing valuable information for the dating and provenance of the manuscript. In addition, these fragments may come from books that are more valuable than the book that hosts them, or they may contain unpublished historical content.

This paper deals with the PT investigation of texts buried in two original ancient books preserved at the Biblioteca Angelica of Rome and the capability to read them. In the following, after briefly describing the main aspects affecting the readability issue and reviewing the peculiarity of the PT signal originating from buried ink elements, we present the experimental results obtained by analyzing texts lying at different depths in historical bookbindings. In particular, we report on the quantitative analysis of the contrast and the distortion, two parameters that affect the quality of the thermographic images of the detected texts and, therefore, their readability. It is indeed the application of this quantitative thermographic approach to the readability analysis of historical buried texts that represents the main novelty of this work.

2. Thermal Reading of Buried Texts: Signal Theory

2.1. Thermography and Texts Buried in Historical Bookbinding

Recently, some studies have been presented where PT was successfully employed to investigate graphic elements lying beneath paper and parchment layers [10,11,12,20]. However, these studies are based on the analysis of specially made laboratory samples and do not include the validation of the proposed method on original manuscripts. On the other hand, PT has been successfully used for the study of historical bookbindings [21,22] with written or printed scraps inserted between the cover and the end-leaf [18,19]. In this respect, PT has proved to be very effective in revealing the texts lying on such scraps that otherwise could no longer be accessed [10]. In addition, many studies carried out on historical artifacts have shown that thermographic images can even allow for their thermal reading, this result being of utmost importance to codicologists and paleographers. These studies have clearly shown that the possibility of obtaining images of a hidden text that are sharp enough to allow it to be read depends on structural characteristics that, so far, have never been studied quantitatively in historical artifacts. In particular, they include the nature of the overlying materials, the interface between these and the text support and, above all, the depth of the text itself from the outer surface.

As an example, Figure 1 shows two thermograms extracted from recorded sequences and processed in a way described later on in Section 3.2. They display buried texts located at different depths in the same bookbinding. In particular, in the thermogram recorded at the earlier delay time (see Figure 1b), the written parts buried at a 95 μm depth are visible, while in the one corresponding to a later delay time (see Figure 1c), the text located at a 155 μm depth also becomes evident, partly overlapping the shallower text but also laterally extending beyond its position. As can be seen, the text shown in Figure 1c does not possess the same degree of readability as the one in Figure 1b due to its deeper position. In fact, as discussed in more detail later on in Section 4, under similar experimental conditions, the features shown by the recorded thermograms become less contrasted and more distorted with the increase in the ink layer’s depth. Actually, besides the decrease in the thermogram contrast, it is found that the text readability is also negatively affected by lateral heat diffusion effects. In fact, even by employing the uniform VIS light illumination of the sample surface, heat diffusion within the sample volume occurs according to a 3D regime when the light absorption is not homogeneous [12].

This is the case of the studied buried texts, where the heat is more efficiently generated at the ink position and diffuses into the colder surrounding material, where the light absorption is less, in the directions both parallel to the sample surface and perpendicular to it. Therefore, the apparent width of the ink strokes that can be deduced from the recorded thermograms increases with the time delay, thus leading to the blurring of the PT images and, consequently, to a more difficult reading of the written parts. As a confirmation of this, thermograms obtained from a given buried text for different values of delay time are shown in Figure 2, where the edges of the ink characters become progressively smeared out because of the role played by the progressive lateral heat diffusion.

With regard to this kind of investigation, it is worth mentioning that alternative techniques employing a mobile macro-X-ray fluorescence (XRF) [23] scanner or computed tomography [24,25] have been quite recently proposed. Such an XRF technique even enables the elemental composition evaluation of fragments located deep in the bookbinding. However, the technique proves to be effective only for the investigation of texts written with iron-based inks and, unlike PT, cannot be applied to carbon-based ones. If one considers that, in order to visualize the buried scraps, it is often necessary to dismantle books that may be of historical relevance, it becomes evident that it is of crucial importance to optimize both the experimental configuration and the data processing method of the involved non-destructive techniques so as to improve the possibility of reading such buried texts. To this aim, a number of models for the PT signal originating from ink features buried between paper layers have recently been presented by the authors of this work [11,12]. Such models have been developed assuming both 1D and 3D regimes for heat diffusion and then tested on specifically designed laboratory samples. In particular, the results in Ref. [10] have proven the capability of PT in detecting texts buried beneath book endpapers. In Ref. [11], the same authors provided a model for PT signal analysis considering a one-dimensional heat diffusion model for the analysis of subsurface ink features. In Ref. [12], the model was extended to account for 3D heat diffusion, particularly for the analysis of the lateral diffusion affecting the quality of the thermal reading. Unlike the aforementioned studies, based on the analysis of ad hoc prepared laboratory samples, the present paper deals with PT applied to real case studies and, in particular, to the analysis of the readability of texts buried in historical bookbindings. In the following, after briefly reviewing the main points related to the PT signal originating from buried graphical features, we report experimental PT results that have been obtained from buried texts located at different depths of the studied historical bookbindings. Thereafter, we present the quantitative analysis of both the contrast and the distortions in the recorded PT images, which are then related to the physical and structural properties of the investigated samples.

2.2. PT Theory

As mentioned above, PT may enable the detection of buried ink features even if their presence induces only marginal changes in the induced temperature distribution in the investigated item. In fact, unlike optically opaque materials, in semi-transparent media, the contrast in the IR emission may also originate from inhomogeneities in optical properties; for example, the VIS light absorption and IR emissivity of the buried ink layers are most often greater than those of the surrounding material.

Based on the considerations reported above, theoretical models for the PT signal originating from ink features buried in a semi-transparent material have been developed and described in detail elsewhere [11,12]. Briefly, according to the schematic representation depicted in Figure 3, in such PT signal models, a uniform sample slab of thickness H is considered, where a thin graphical element is located at depth d. As a result of the VIS light absorption taking place both within the sample slab and at the buried ink feature, a time-varying 3D temperature distribution $T\left(\overline{x},z, t\right)$ is induced, where $\overline{x}$ and z are directions parallel and orthogonal to the sample surface, respectively (see Figure 3).

As regards the PT signal $S\left(\overline{x},t\right)$, the IR radiation emitted from the sample volume must be taken into account, and, therefore, the following expression for the signal is obtained [11]:

$$S\left(\overline{x},t\right)=K\left[{{\displaystyle \int }}_0^d\beta T\left(\overline{x},z,t\right)e^{-\beta z}dz+\eta T\left(\overline{x},t\right)e^{-\beta d}+{{\displaystyle \int }}_d^H\left(1-\eta \right)\beta T\left(\overline{x},z,t\right)e^{-\beta z}dz\right]$$

(1)

where K is a constant accounting for different experimental factors, like the VIS light intensity, IR radiation detection efficiency and other geometrical aspects; β is the effective value of the material’s IR absorption coefficient over the detected wavelength range; and η is the ink layer’s IR absorptance. It is worth noting that the expression for the signal originating from ink-layer-free areas is equivalent to the first integral evaluated from 0 to H.

In Figure 4, the expected PT signal profiles $S\left(\overline{x},t\right)$ over the edge of a buried ink feature are shown. According to the 1D diffusion regime, where only the heat diffusion along the direction orthogonal to the sample surface is considered, one would predict a sharp change in the signal profile, as sketched by the dotted line, because of the greater IR radiation emission from the buried drawing with respect to that from the weaker-absorbing surroundings. However, as the ink layer gets heated by the VIS light transmitted through the overlying paper layer and by the temperature field diffusing from the surroundings, the heat also begins to propagate from the ink toward the cooler regions along the $\overline{x}$ direction, thus leading to a smoothing of the temperature profile over the edge, as outlined by the gray plot in Figure 4. In order to account for the characteristics of the PT signal profile over the ink edges, two specific parameters of the recorded thermograms have been introduced. With reference to the notation adopted in Figure 4, the first one is the contrast $C=S_{max}-S_{min}$, which is defined as the difference between the PT signal values detected over the buried ink feature and where no ink is present, respectively. It is clear that the text readability is ensured, among other factors, by large enough values of C. The second parameter is the distortion index, ∆, corresponding to the distance between the values of $\overline{x}$ where the signal differs by a detectable fraction (2%) from its maximum and minimum measured values: $\Delta =x_{max}-x_{min}$ [12] (see Figure 4). Basically, the distortion index measures how much the detected $S\left(\overline{x},t\right)$ profile differs from the step-like one, where ∆ = 0.

In Figure 5, the time dependence of both the contrast and the distortion index following the onset of the heating pulse are shown. In these simulations, both the optical and thermal properties of the paper leaf have been assumed to be equal to those typical of paper (α = 1.3 × 10² cm⁻¹; β = 2.4 × 10² cm⁻¹; D = 1.0 × 10⁻³ cm²/s) [11], with those of the ink equal to those of paper but with an absorptance η = 1. As can be seen in Figure 5a, the contrast reaches its maximum value after a typical delay time that increases with the ink depth. This is so because of the time needed by the heat to diffuse from the shallow volume where the VIS light is absorbed to the ink-layer position where the IR emissivity is the greatest. At later times, a decrease in C(t) can be observed because of both the ongoing heat diffusion past the ink-layer position and also along the lateral direction and the heat losses from the sample surface. In addition, Figure 5a shows that the maximum value of the contrast is reduced as the ink depth increases since the ink layer becomes less effectively heated due to the attenuation of both the heating VIS light and the temperature field reaching the layer because of the larger distance involved. An additional factor is the greater attenuation of the emitted IR radiation from the buried graphical feature during its propagation through the overlying paper layer before it emerges from the paper surface. In addition, the distortion index (see Figure 5b) grows monotonically with the delay time due to the corresponding increase in the lateral heat diffusion in the resulting PT signal profile, and, at equal delay times, it turns out to be the same irrespective of the depth at which the ink layer is located. This is due to the fact that the index is solely dependent on the time during which lateral diffusion occurs, a process governed by the value of the thermal diffusivity of the material. On the basis of the plots reported in Figure 5, it may be concluded that the effective readability of a given subsurface graphical feature is crucially dependent on the depth of its position. In fact, with the increase in depth, the maximum contrast is both decreased in amplitude and shifted to later delay times, with a consequent increase in the lateral diffusion and, therefore, in the distortion in the recorded thermogram.

3. Samples and Experimental Setup

3.1. Samples

Thermographic investigations were performed on two original ancient books preserved at the Biblioteca Angelica of Rome. The first one is a book printed in 1758 (ʘ.2.16) with a board parchment binding and uncolored endpapers (Figure 1a). Its binding is characterized by having two superimposed spine wings sandwiched in between the endpapers and the boards. The wings were made from scraps of reused paper, both written on one side. The cross-section of this structure is sketched in Figure 6a. The second book was printed in 1592 (f.9.31). It has a limp parchment binding with uncolored endpapers (Figure 2) attached to written spine wings with the text at the outermost surface (Figure 6b).

3.2. Experimental Setup

The measurements were carried out by means of the PT setup described hereafter. Sample heating was induced by means of two flash lamps of 3 kW maximum electrical power, oriented at 45° with respect to the sample surface, delivering pulses a few ms long. The used flash lamps allow the light power to be adjusted to achieve an adequate signal-to-noise ratio without overexposing the artifact to excessive amounts of radiation. In addition, their light pulses were short enough to allow the application of the adopted theoretical model to the analysis of the experimental data. The emitted light from the lamps was filtered in order to remove the IR spectral component and, consequently, to suppress spurious contributions to the PT signal originating from reflections of IR radiation at the sample surface. The detection of the IR radiation emitted by the sample was carried out by means of a Cedip JADE MWIR camera (320 × 240 pixels, InSb focal plane array, 30 μm pitch, 3.6–5.1 μm wavelength range), which is characterized by a Noise Equivalent Temperature Difference (NETD) < 25 mK at 30 °C. The Altair 5.50 software was used to process the images, which were recorded with a 150 Hz frame rate. Thanks to the above-mentioned specifications, the IR camera employed provides a sufficient spatial resolution and frame rate to obtain spatial and time sequences with an adequate amount of experimental data and a sensitivity in the operating spectral range high enough to ensure an adequate signal-to-noise ratio. In order to enhance the contrast in the recorded thermograms, the frame obtained just before the heating pulse was subtracted from all subsequent ones constituting the recorded sequence.

4. Results and Discussion

Figure 7a–c shows the thermograms of text I (see Figure 6) buried just beneath the endpaper in the blue-framed area I, previously shown in Figure 1b, obtained for increasing delay times 0.02 s, 0.05 s and 0.30 s with respect to the heating light pulse. Figure 7d reports the PT signal profiles obtained from the three thermograms over the ink edge of a selected letter. The contrast of the thermographic signal shows similar values for the first two delay times, followed by a subsequent decrease with further delay, while the range of the distortion region (delimited by the vertical bars), corresponding to the distortion index value, increases continuously. Similarly, in Figure 8, the thermograms and the corresponding PT signal profiles, obtained at the same values of delay times used for the data in Figure 7, correspond to the area of a letter of the deeper text previously shown in Figure 1c, corresponding to the red-framed area II in Figure 1a (text II in Figure 6a).

Given the increased depth of its position with respect to that displayed in Figure 7, the contrast of the text is very low at short delay times; it then reaches a maximum and then decreases with increasing delay times, with the maximum occurring at later times with respect to that in Figure 7. The distortion index increases with the delay time in the same way as in Figure 7.

Finally, Figure 9 reports the corresponding results obtained over a letter of the text displayed in the black-framed area (III) in Figure 2. The values of the contrast do not show substantial changes with increasing delay, while the distortion index correspondingly increases. Concerning the depth at which the three different analyzed texts are buried, the following observations should be made. In the case of volume ʘ.2.16, the depths of texts I and II are unknown, and one can only observe that text II is located at a depth larger than that of text I, as witnessed by the longer time required for it to appear in the thermogram sequence. This is consistent in the cross-section sketches in Figure 6a, where the depth of text II is considered to be that of text I plus the thickness of the shallower wing placed on top of the deeper wing. Regarding text III of volume f.9.31, the presence of some tears and gaps in the guard sheet allowed its thickness to be measured with a micrometer, obtaining 105 ± 5 μm.

This thickness should, however, be considered only approximate, as the effective thickness may be affected by the presence of the glue layer and possibly also by the mechanical effects caused by the pressure exerted during the gluing process.

Finally, in Figure 10a,b, we summarize the significant information gathered from the results discussed so far as to the influence of the depth of the investigated texts and the corresponding delay times required to achieve their best thermographic readability. In particular, in Figure 10, we report the experimental results obtained from the previous graphs (open symbols) displayed on the corresponding theoretical (solid lines) and experimental (solid symbols) curves of the delay-time dependence of both the contrast and distortion indexes obtained from the thermogram sequences of the three analyzed samples.

The continuous lines were obtained by means of Equation (1) using values for the optical and thermal parameters typical of paper, previously referred to in the theory section, and effective paper thickness values that would yield the best agreement with the experimental data. The solid symbols in Figure 10a correspond to the differences between the PT signals recorded over the letter analyzed in the plots in Figure 7, Figure 8 and Figure 9 and a non-inked reference area. The following conclusions can be drawn:

• The excellent agreement between the theoretical values and the experimental data confirms the soundness of the proposed theoretical model for the PT signal and its capability to provide reliable indications concerning the values of the depth of the buried ink layer once the values of the optical and thermal parameters of the involved paper layers are known, as well as the parameters affecting the readability of the buried texts.
• With the increasing depth of the buried ink layer, the peak contrast value decreases, and its occurrence is shifted to longer delay times.
• The time dependence of the distortion index shows that it is related only to the delay time of the detection of a particular subsurface text, irrespective of its depth, as confirmed by the superposition of the values corresponding to texts buried at different depths of the theoretical curves and the corresponding experimental data reported in Figure 7, Figure 8 and Figure 9.

Therefore, concerning the readability of the different texts, the following can be concluded:

• For a relatively shallow text (95 μm deep), the best readability is obtained in the earliest thermogram (Figure 7a) because of the highest achieved contrast and smallest distortion.
• For the deepest text (155 μm), the largest contrast can be obtained at longer delay times (Figure 8b,c), thus compromising the possibility of optimal readability because of the correspondingly increased distortion.

5. Conclusions

In the present study, pulsed thermography was applied to the detection of text buried below the end-leaves of ancient manuscripts of significant historical relevance. A quantitative analysis of the obtained thermographic results was carried out by using a theoretical model to elaborate on the quantities actually affecting the readability of the text retrieved in the thermograms at different delay times with respect to the pulsed heating source.

In particular, thermogram sequences, as a function of the delay time, of texts buried at different depths below the end-leaves of different ancient books housed in the Biblioteca Angelica in Rome were recorded. Two parameters were introduced, the contrast index and the distortion index, and they were used to evaluate the degree of readability of the texts retrieved in the thermograms. The contrast dependence on the delay time showed a peak feature, whose value decreased with the increasing depth of the buried ink layer and moved to greater delay times. Moreover, the distortion index was shown to be related only to the delay time of the detection of a particular subsurface text, irrespective of its depth. These results were confirmed both by theoretical calculations and by experimental results, with excellent agreement between them.

The depth of the subsurface features has been found to play a crucial role in the determination of the retrieved texts’ quality. For relatively shallow texts, the best readability is obtained in thermograms recorded at early delay times because of the sufficiently high achievable contrast and small distortion. For deeper texts, sufficient contrast can only be obtained at longer delay times, producing a larger distortion and therefore compromising the possibility of optimal readability.

Based on the consistency of the experimental method and the reliability of the proposed theoretical model, which closely reflected the results obtained experimentally, the presented approach makes pulsed thermography a very reliable method for the study of texts buried beneath paper sheets within the bookbinding of ancient books.

On the basis of the results reported in this work, in the near future, research efforts should be specifically devoted to the design of algorithms for the processing of PT images in order to improve the readability of the detected hidden texts, including procedures involving artificial intelligence. Moreover, further research activity should also be dedicated to the development of integrated approaches where PT is employed in combination with other possible imaging techniques in an effort to obtain complementary characterizations of the subsurface graphical features.