Photothermal approach for cultural heritage research

Cultural heritage items are extremely valuable objects and, over the last few decades, they have been the subject of intensive scientific investigations. Such studies are mainly driven by the need to gather useful information concerning, in particular, their state of conservation so as to optimize their conditions of preservation. Among other things, this involves both an evaluation of the structural integrity of an artwork and an identification of the constituent materials in order to help restorers in the preservation of artwork items. Moreover, such investigations have also been proven helpful in getting a better insight into the manufacturing process of an artwork, such information being of great importance for scholars, in particular. In this respect, it is worth noting that some of the most valuable information is often provided by features that lie beneath the surface layer which, in most cases, is not accessible through ordinary optical inspection. Consequently, over the last few decades, significant efforts have been made to develop experimental techniques for a nondestructive evaluation of subsurface features in artworks.

The techniques adopted for analyzing artworks can be divided into two main categories. The first one includes techniques aimed at evaluating the chemical composition of the constituent materials, for example, X-Ray Fluorescence (XRF)^1,2 and optical spectroscopy. The second category refers to methods that allow an inspection of the internal structures of objects, for example, x-ray radiography³ or terahertz (THz) imaging.⁴ The present paper will focus mainly on the latter kind of applications in which photothermal techniques have been proven to be very effective. In fact, artwork can generally be imagined as multi-layer structures and experimental techniques that are able to provide depth-resolved characterization are considered to be of great relevance in the field of cultural heritage investigations. Most conventional techniques that are able to reveal subsurface features provide only two-dimensional projections of three-dimensional structured objects, thus making it difficult to depth-resolve the detected items. In this respect, photothermal techniques^5–7 can prove effective by probing the time-varying temperature distributions that are stimulated in samples by the absorption of either pulsed or periodically modulated light. It has been shown that, in the case of periodic heating, the resulting temperature field may be mathematically represented in terms of thermal waves.⁸ In the following, we will refer even the temperature fields produced by different kinds of time-dependent heating sources as thermal waves, such as the cases of pulsed or step-like heating, meaning to identify the propagation of the thermal front produced by the time-dependent heat diffusion. The evolution of such thermal waves will depend not only on the amount of the absorbed light but also on the sample thermal properties and geometrical characteristics. Therefore, an analysis of the behavior of thermal waves, by means of one of the several existing photothermal techniques differing from each other basically for the scheme adopted for their detection,⁷ can prove very effective for gathering information on the sample optical, thermal, and structural properties. As shown in the next paragraphs, such a possibility has also been exploited in the field of cultural heritage investigations where the characterization of both the physical and structural properties of the constituent materials of artwork can be of great relevance.

Besides being used for the determination of material properties, over the last few years, photothermal techniques have been increasingly used for the detection of subsurface features.^9,10 Here, the features refer to relatively small regions characterized by different physical or structural properties with respect to those of the surrounding material. The presence of such features affects the resulting local thermal wave propagation within a sample and, consequently, the correspondingly detected photothermal signal, hence making it possible to carry out depth-resolved quantitative characterizations. In fact, the penetration depth of thermal waves into samples can be controlled by either varying the modulation frequency in the periodic heating regime or letting the heat diffuse over different time durations in the case of pulsed or step-heating schemes. In all cases, photothermal signals can carry information about the local sample properties over the depth reached by the thermal wave. Therefore, by selectively varying such a penetration depth, it is possible to obtain quantitative information such as the depth or the lateral size values of the detected subsurface features.

Among the various photothermal techniques, Infrared Thermography (IRT)¹¹ has been proven to be very effective for the investigation of cultural heritage items, thanks to its remote generation and detection capabilities, relatively simple operation mode allowing in situ investigations of artwork, and the possibility of inspecting large sample areas over relatively short times. As shown in the following, IRT provides 2D images enabling a stratigraphic study of artwork, thanks to the possibility of varying the thermal wave penetration depth.

In the following, after reviewing the operating principles of the photothermal techniques and, more specifically, of IRT, situations that are more typically encountered during inspections of cultural heritage items are described. For the sake of brevity, the following presentation will be limited to the artwork categories that have been more extensively investigated, i.e., paintings, bronze statuary, books, and documents. For each category, the main features of the artwork structures are presented and, based on this, the specific objectives of the IRT investigations carried out on them are discussed. Finally, we report on the recent progress that has been made in the application of photothermal techniques to the field of cultural heritage and on some future perspectives. In fact, despite the successful investigations that have been carried out so far by means of IRT, a lot of efforts are currently being made to improve the effectiveness of the IRT analysis. The reason for this is that a limited amount of optical energy must be used for the thermal stimulus of the investigated items so as to avoid damage associated with excessive light absorption and large temperature rise. In this respect, there is an ongoing work aimed at the development of data processing techniques, which will result in an enhancement of the detectability of the subsurface features under conditions of poor signal-to-noise ratio. On the other hand, integrated multidisciplinary approaches are attracting increasing attention from researchers. In these approaches, IRT is applied in combination with other inspection techniques that are able to provide complementary results with those of IRT, thus allowing more accurate in-depth studies of artwork.

As mentioned earlier on, photothermal techniques rely on the detection of thermal waves induced by the absorption of radiation whose spectra are mostly in the visible (VIS) range. Pulsed or periodically modulated heating light beams are most commonly used even if other kinds of heating schemes can be adopted, as shown in the following. In this paragraph, the main properties of the thermal waves are discussed and, for the sake of brevity, such a presentation is limited to a few but significant cases, while a more detailed discussion can be found Refs. 5–8. In particular, we will refer to the case where the induced heating is assumed to be uniform over the sample surface so that the thermal waves diffuse according to a 1D regime along the direction orthogonal to the heated surface. To analyze the effect produced by the light absorption, it is convenient to distinguish between the cases of optically opaque and optically semi-transparent samples. With regard to cultural heritage artwork, it is intuitive to expect that, for instance, in bronze statues, the absorption of the heating beam takes place at the sample surface, while in paper documents, it occurs throughout the sample volume. Owing to such a circumstance, the same heating scheme, in the two different sample categories, can give rise to thermal waves characterized by very distinct properties, which must be taken into account in the analysis of the obtained photothermal signal.

In the case of semi-infinite, homogeneous, and optically opaque samples, the induced temperature rise T(z,t) (where z is the in-depth coordinate) shows an exponential decay as it propagates into the sample volume. In particular, if the sample heating is produced at instant time t = 0 by light pulses whose duration is much shorter than the other characteristic time involved in the heat diffusion, the induced thermal waves amplitude decays to 1/e of the surface value over a depth $μ= 2 D t$ (where D is the sample thermal diffusivity), known as the thermal diffusion length, which can be considered as the effective penetration depth of the induced thermal wave. From the expression of $μ$⁠, it becomes evident that, by analyzing the temperature response obtained at different time delay values following the heating pulse, it is possible to probe the sample interior over varying depth values. In addition, the temperature at the sample surface T(0,t) decays with time as $1/ t$⁠, regardless of the sample thermal properties. Therefore, deviations from this time dependence may indicate that the sample is either inhomogeneous or that the sample thickness is finite, thus providing the possibility of gathering information about the sample structure or, alternatively, of evaluating D once the sample thickness is known.¹² 

In the periodically modulated heating regime, thermal waves oscillating at the same frequency f of the heating beam are induced which, when using the phasor formalism, can be conveniently expressed as $T( z , t)=Re[ ( T ~ ( z ) ) e j 2 π f t]$⁠. Here, Re stands for the real part of the complex expression, j is the unit imaginary part, and $T ~(z)$ is the complex temperature amplitude. Therefore, the photothermal signal is comprised of both the amplitude and phase of $T ~(z)$ whose frequency dependence allows one to gather information about the investigated samples. Similarly to the results found in the pulsed heating regime, in the case of optically opaque samples, the amplitude $| T ~ ( z )|$ decays exponentially over the depth z where the corresponding thermal diffusion length is $μ= D / π f$⁠. Therefore, by varying the modulation frequency, it is possible to change the thermal wave penetration depth into the sample. Moreover, at the sample surface, the amplitude of $T ~(0)$ follows the $1/ f$ dependence, while its phase displays a constant value equal to –45°. As already discussed in the pulsed heating case, many approaches for the non-destructive evaluation of samples rely on the deviations of the observed photothermal signal from this expected behavior.

As regards optically semi-transparent samples, the absorbed light intensity I(z,t) is typically assumed to decay into the sample according to the Beer–Lambert law as I(z,t) = I₀(t)exp(−βz), where β is the mean optical absorption coefficient in the heating beam spectral range, while I₀ is its intensity. In both pulsed and periodically modulated heating regimes, thermal waves are generated over a depth of the order of 1/β due to the direct heating produced by the absorbed light. Therefore, in semi-transparent samples, it is more difficult to discriminate between features located at different depths due to the instantaneous sample heating occurring simultaneously over the entire irradiated sample volume which makes the selective depth analysis of the thermal waves more complicated.

So far, in the presentation of the main properties of the thermal waves, only the heat diffusion taking place in semi-infinite homogeneous samples has been considered. However, it is quite evident that if such conditions are not met, the behavior of the induced temperature field can become significantly different with respect to the one described above. The influence of inhomogeneities inside the sample volume on the resulting thermal wave field depends on several factors like the lateral size, the depth, and the difference in its physical properties as compared to those of the surrounding material. As regards the thermal properties, the thermal effusivity $e= ϱ c k$⁠, where $ϱ$ is the density, c the specific heat, and k the thermal conductivity, plays a crucial role.^7,9 In fact, thermal waves interact with the subsurface features in a manner that is mathematically equivalent to the reflection process of conventional waves, being the reflection coefficient dependent on the difference between the thermal effusivity value of the feature and that of the surrounding medium. Thus, any feature located beneath the surface within a thermal diffusion length reflects the thermal waves and, hence, becomes visible, provided that its size is large enough to significantly affect the heat diffusion process. Many of the photothermal applications in the field of non-destructive evaluation deal with the detection of subsurface air-filled defects such as voids or delaminations. In the case of pulsed heating, the temperature decays at a lower rate at the surface above the defect due to the poor thermal properties of air, thus giving rise to a thermal contrast given by the difference between the signal recorded over the defective area and that recorded over a sound one. By analyzing the time dependence of the contrast, it is in principle possible to also obtain quantitative information about the defect. The same considerations can also be extended to the periodical heating scheme where the photothermal signal detected over a subsurface feature shows a different dependence on the modulation frequency with respect to that corresponding to a homogeneous part of the artwork.

Features characterized by different optical properties with respect to the surroundings can also affect the resulting thermal wave. In fact, the presence of locally different values of the optical absorption coefficient can lead to a nonhomogeneous heating of the sample and, hence, to a temperature distribution showing different features with respect to that induced in an optical homogeneous sample. Differences in the local infrared emissivity can also play a crucial role in the case of photothermal techniques based on the detection of the emitted infrared radiation.

As mentioned above, photothermal techniques enable the detection of subsurface features, provided that these are located within the thermal wave propagation depth. The thermal diffusion length can be tuned by appropriately varying the modulation frequency or the delay time. However, by doing so, the photothermal signal becomes unavoidably affected by attenuation or other detrimental factors, like heat losses from the sample surface or 1/f noise, up to the point where deeply buried features become barely visible in the raw data. In an effort to overcome such difficulties, over the last few years, several studies have been devoted to the development of algorithms, which is still ongoing, for the post-processing of photothermal signals with a small signal-to-noise ratio to enable the detection of deeply buried subsurface features, as described later on.

Among the different photothermal techniques, IRT is the one most widely used for the study of cultural heritage items.^13–15 IRT is based on the locally resolved recording of the time dependence of the emitted infrared (IR) radiation following the sample heating. The recording is accomplished by means of an IR camera that provides images called “thermograms.” In IR optically opaque samples, the amount of emitted radiation is proportional to the surface temperature and, therefore, the thermograms correspond to the temperature distributions at the sample surface. On the contrary, in IR semi-transparent samples, it is necessary to also take into account the IR radiation emitted from the sample volume and, consequently, the IRT signal becomes dependent on the induced temperature distribution within the sample.^16,17 As discussed in the next paragraphs, different approaches must be adopted for the analysis of the IRT experimental data obtained in IR opaque and semi-transparent artwork, respectively, enabling the retrieval of different kinds of information.

Among the different IRT configurations, pulsed thermography (PT) may be considered as the simplest and quickest way of carrying out thermographic inspections since it probes the transient variation in the emitted IR radiation following the pulsed heating of the sample. In the case of periodically modulated heating, the lock-in thermography (LT) detection scheme is adopted to obtain the images corresponding to the signal amplitude and phase induced by the emitted IR radiation component oscillating at the same frequency of the heating beam. As with other PT techniques, the presence of subsurface features leads to a contrast in the thermal map of the sample surface with frequency-dependent characteristics.

As sketched in Fig. 1, the active thermography setup basically requires an optical source providing the heating light beam, an IR camera for the thermogram recording, and a computer for the post-processing of IRT data. As regards the optical sources, flashlamps are adopted to induce pulsed heating, while light projectors or LED arrays are used in the periodical modulated heating schemes.

Thermographic measurements are typically carried out in the IR transparent atmospheric windows’ ranges corresponding to the Medium-Wave Infrared (MWIR) (∼3–5 μm) and to the Long Wave Infrared (LWIR) (∼8–14 μm) spectral ranges. For equal temperature difference between an object and its background, it is shown in the literature that the thermal contrast is approximately twice as large in the MWIR range as compared to the LWIR one.¹⁸ The detection is accomplished by means of an IR camera based on photonic detectors mainly made of HgCdTe and InSb. IRT investigations can be carried out in either the reflection or the transmission scheme. In the former case, both heating and IR recording take place at the same sample surface, while in the latter, the IR radiation emitted from the surface opposite to the heated one is detected. With regard to the investigation of cultural heritage items, reflection geometry is very often adopted since, in almost all cases, artworks can be conveniently accessed from only the external surface. In addition, such geometry offers better sensitivity to the presence of subsurface features compared with that of the transmission one.

IRT investigations of subsurface features in cultural heritage items rely on the contrast that they are capable of inducing in the recorded thermograms. However, spurious effects such as those associated with inhomogeneous heating, optical reflections, and emissivity variations at the sample surface can give rise to additional contrast in the thermograms superimposing on that produced by the internal features, thus making the analysis of the recorded images more complex. To overcome these kinds of problems, several postprocessing procedures have been proposed. For example, a pixel-by-pixel subtraction of the frame obtained just before the heating pulse from all the thermograms in the sequence after the pulse has been proposed in order to remove static background contributions to the signal and to increase the signal dynamics. A normalization procedure can then be carried out, which consists of the division of the signal from each pixel by its corresponding initial value observed during the sequence to eliminate possible IRT signal variations arising from inhomogeneities in optical properties.

In order to further address the problem of the spurious contrast in the IRT images and also to enhance the visibility of the subsurface elements, several additional kinds of signal processing procedures have been proposed whose development is in many cases still in progress. Here, we will limit ourselves only to a very brief description of their basics.

Among the first to be introduced, it is worth mentioning the differential absolute contrast (DAC) technique¹¹ where the mean sample surface temperature value, with respect to which the features’ thermal contrast is obtained, can be calculated even in the case where no homogeneous sample area is available. In the thermographic signal reconstruction (TSR) technique,¹⁹ the time derivative of the recorded image sequence is computed to identify the occurrence of PT signal deviations from the $1/ t$ dependence. As pointed out earlier on, in an optically opaque sample, such a circumstance may indicate that the propagating heat front has reached the sample surface opposite the heated one or the presence of thermal inhomogeneities within the inspected sample, thus providing information about the sample’s internal structure. TSR has proven to offer an improved detection sensitivity and also to be less affected by spurious signal contributions arising from non-uniform heating or optical properties’ variations at the sample surface. In the pulsed phase thermography (PPT) technique,²⁰ the pulsed thermographic signal is Fourier analyzed to obtain the amplitude and phase image of the oscillating temperature contributions of the different frequency components. One of the main advantages of PPT is that, in opaque samples, the phase images are less affected by non-uniform heating and emissivity conditions than the amplitude ones and they also turn out to show better sensitivity to defects located at relatively large depths.

Recently, IRT image analysis based on the statistical processing of raw data has attracted increasing attention by researchers. The first method to be introduced was the principal component thermography (PCT) in which the IRT collected data are represented as a superposition of orthogonal thermal images, known as principal components.^21,22 Such components are obtained in such a way that the first component corresponds to the most common temperature distribution over the sample surface, while the subsequent ones take into account deviations from this common trend due to the presence of subsurface features. In this respect, PCT offers the advantage to compress information into a few images since only a few of the first few components are of interest for the sample evaluation, while the subsequent ones basically correspond to trivial or noise contributions. In order to reduce the computational complexity which can become relevant with an increase in the number of frames, the Candid Covariance-Free Incremental Principal Component Thermography (CCIPCT) has been proposed,²³ which uses a faster computational procedure to estimate the principal components.

One of the main drawbacks of PCT is that the images consist of a superposition of features located at different depths into the sample volume, thus making the evaluation of their depths less straightforward. However, based on the successful applications of PCT, research efforts are currently focused on the development of other statistical methods for the analysis of thermal images. Among others, it is worth mentioning the approach that makes use of the Partial Least Squares Regression (PLRS) method to obtain an image sequence that accounts for the most important features while irrelevant contributions are ignored.²⁴ 

In the following, an overview of the results obtained in the study of cultural heritage by IRT is presented by grouping them according to the kind of investigated items, namely, paintings, bronze statuary, and books.

IRT has been proven to constitute a valuable tool for a non-invasive analysis of paintings,^25–28 and current research is focused on further improving its effectiveness. As shown in the following, such improvements involve the design of both new experimental configurations and data processing techniques and also the development of multidisciplinary approaches where IRT is used in combination with other techniques that can provide complementary information.

In the investigation of paintings, IRT is almost always applied in the reflection configuration since it is more sensitive to the features located into or right below the painting layers and, in addition, it does not require access to the artwork rear side which, in many cases, may be very difficult to obtain. In this respect, the use of robotized systems enabling both the scanning of the heating light beam and the subsequent IRT image recording over the painting surface has been recently designed, such systems being especially useful for the analysis of large paintings.²⁹ In fact, owing to the limited spatial resolution of the IR images which up to recent times could be achieved with MWIR cameras, IRT investigations very often require the recording of the images of different painting sub-sections, ensuring the adequate required resolution, and they are then assembled into a mosaic image. In particular, in Ref. 29, the line-scan thermography (LST) system has been proposed. Such a system makes use of a pulsed line source operated by a robot to scan-heat the whole painting surface, while the IR camera is moved synchronously to record the corresponding IR signal. The IR image is then reconstructed by mapping together the results of the scanned signal.

Panel paintings can be considered a multi-layer structure consisting of a ground support, on top of which paint and varnish layers are laid. According to the manufacturing procedures followed during the 13th and 14th centuries, the background is made of a canvas fixed on wooden support. On the canvas surface, a ground layer typically made of gypsum is applied to provide a smooth surface on which to execute the painting. The adopted paints consist of fine grain pigment powders dispersed into a binder. Finally, a transparent varnish layer is applied whose role is to both protect the painting and sometimes modify the perception of colors. As shown later on, information obtained on each component of the painting structure may be of great importance.

One of the primary goals of IRT investigations is to detect defects and repairs that may lie inside a painting structure. The presence of defects is often related to the different deformations that the painting layers undergo, associated with the daily fluctuations of the ambient parameters, and to the different elastic properties of the layers. This can possibly lead to the production of cracks as the layers become less flexible with time. In particular, such deformations are responsible for the formation of craquelure, namely, cracks in the paint layers and uplifts due to the loss of adhesion between the different layers of the painting structure. Additional factors, such as biodeterioration by fungi or the presence of larvae and worms producing long holes into the wooden support, may also cause damage to the painting.

PT has proven to be an effective tool for the relatively speedy detection of the above-mentioned defects in both specifically fabricated test samples and original artworks. In order to further improve PT capabilities, great efforts are currently being made to develop and subsequently apply post-processing algorithms enabling the detection of the subsurface features that are barely visible in the raw PT data. These include the above-mentioned DAC, PPT and TSR, and, more recently, the ones based on statistical analysis such as PCT and PLSR which have been shown to be very effective in improving the visibility of the defects.³⁰ 

One of the drawbacks of the PT application in the investigations of paintings may be related to the need to use relatively high-energy heating pulses in order to ensure an adequate signal-to-noise ratio for the detection of the subsurface features. However, excessive temperature changes may result in thermal stresses which can eventually lead to damage. In addition, color changes in the pigments can also be induced due to irreversible thermochromic and photochromic effects, thus altering color perception. To overcome these disadvantages, over the last few years, pulse compression thermography (PuCT)^31–33 has been proposed whose development is still in progress. PuCT makes use of low-power light sources that are modulated with time-varying frequency in order to produce heating profiles, in which, unlike in ordinary PT, the bandwidth and the duration become uncorrelated. Therefore, the duration of the heating pulse can be increased to improve the signal-to-noise ratio, while, at same time, the bandwidth can be tuned in order to get the desired thermal wave penetration depth. Thanks to such a procedure, it is possible to successfully carry out investigations by inducing a moderate temperature rise (∼1 °C³³ over the sample surface, and basically independent frequency and time domain analysis of the IRT signal can be achieved. As mentioned earlier on, research efforts have recently been directed toward the development of multidisciplinary approaches in which IRT is used in combination with other techniques that can provide complementary information about artworks. This is the case with painting investigations carried out by the combined use of holographic interferometry (HI) and IRT.^34,35 HI involves the superposition of two holograms acquired before and after moderate mechanical loading of a painting, respectively. The hologram superposition results in the formation of fringe patterns that indicate the presence of surface displacements due to the presence of subsurface defects such as delaminations, cracks, and voids. Other optical coherent techniques based on the same operating principle, such as shearography, speckle decorrelation, and electronic speckle pattern interferometry, have also been successfully applied in combination with IRT.³⁴ In conclusion, the large detection sensitivity offered by interferometric techniques, combined with the simultaneous quantitative evaluations obtained by means of PT, constitutes a powerful tool for the investigation of paintings.

As regards canvas, deformations, known as cusping, may occur near their edges due to the mechanical tension from the tacks that hold them to the wood support. An analysis of the cusping pattern, which is typically carried out by means of x-ray radiography, is important since it may provide significant information about the painting manufacture and, in particular, about the adopted procedure for the stretching of the canvas. In this respect, the effectiveness of the deformations characterization has been recently improved by combining IRT with CW (continuous wave) terahertz (THz) imaging carried out in both the reflection and transmission configurations.³⁶ Such an approach takes advantage of the ability of THz radiation to probe features located deep into the painting structure since most materials, with the exception of water and metals, are highly transparent in the THz range.

Besides other integrated approaches such as those combining the use of IRT with colorimetry³³ and ultrasound C-scan technique,³⁷ it is worth mentioning the one in which both IRT and Infrared Reflectography (IRR) are applied to the inspection of paintings.^38–41 As explained below, such an approach is attracting increasing attention from researchers due to the further improvements being made, which can lead to interesting perspectives. IRR⁴² is a well-established technique that makes use of the optical contrast in the images (reflectograms) obtained by recording the reflected component of the radiation illuminating the object. Here, radiation in either the near-infrared (NIR) (0.7–1.0 μm) or the short-wave infrared (SWIR) (1–3 μm) spectral ranges is commonly adopted. In fact, thanks to the transparency of the paint layers in both the NIR and, to a greater extent, the SWIR ranges, the radiation can penetrate more easily than that corresponding to the VIS part of the spectrum, and, hence, it can reach the ground where it can be either reflected back by the support or absorbed by the darker features of the drawing.

Thanks to such a circumstance, IRR has been widely used in paintings for the detection of underdrawings on the chalk substrate and/or pentimenti associated with an artist’s change of mind during the realization of a painting. In addition to the underdrawing/pentimenti detection, IRR also provides an effective tool for the visualization of subsurface-like features located into the paint layers, such as the presence of filled areas or retouches.

Such a possibility is granted by the different NIR reflectivities of the pigments that were used to fill the paint voids with respect to that of the surrounding original material.

Figure 2 reports the results reported on a test sample, the portrait of a Madonna in which underdrawings had been first performed and artificial defects introduced at the canvas-support interface.²⁷ 

While the photographic image shows the surface of the paint layers, the IRR clearly shows the presence of the underdrawings due to the larger penetration depth of the IR radiation, and the PT image detects the presence of the subsurface defects at the Madonna's neck and face, thanks to the induced thermal wave diffusing down to the support.

More recently, the possibility of carrying out IRR in the medium-wave infrared (MWIR) (3–5 μm) spectral range has also been exploited.⁴³ In particular, it has been observed that IRR carried out in the MWIR range can allow the visualization of complementary features with respect to those detected in the short wavelength ranges due to the different optical properties that many pigments may show in the different IR ranges and, also, to the larger penetration depth of MWIR radiation compared with that in the NIR and SWIR ranges.

On the basis of the considerations reported above, a comparative analysis of the IRR and IRT images may help probe the paintings over different depths and, in addition, to reveal complementary features whose combined analysis would allow a better understanding of the investigated artwork. As a future perspective, it is worth mentioning that IRR and IRT inspections are normally performed by using separate cameras and setups. However, quite recently, IR cameras operating in the 1.5–5 μm range with separate filter options have become available, thus making it possible to carry out both IRR and IRT imaging by means of the same camera. Such a circumstance should result in a significant simplification in the implementation of the measurements and, in addition, in a better interpretation of the results due to the loss of ambiguity in the spatial correspondence of the features shown by the images obtained in the different IR spectral ranges.

As an example of the combined use on an original artwork of IRT and IRR in the MWIR range, using the same camera, in Fig. 3, the results obtained in the case of the painted altarpiece representing the Virgin and Child, hosted in the basilica of S. Maria in Cosmedin in Rome are given. Figure 3(a) shows the picture of the child, while Figs. 3(b) and 3(c) show the corresponding IRT and IRR images, respectively, which are able to show features that lie beneath the pictorial layers. For example, the darker outline of the right-hand features in Fig. 3(c) corresponds to the underdrawing performed by the artist, as it appears slightly different from the corresponding outline displayed in the picture. Figures 3(b) and 3(c) also show that the thumb profile is missing in the subsurface layers (it has been repainted by the restorer) because of damage that had occurred on the painting substrate in that area. The extended area of such damage is detectable both by IRT and IRR. In the case of IRR, the contrast is due to local differences in the optical properties of the filling performed to restore the damage before the repainting is performed, while in the case of IRT, contrast arises because of differences in the thermal properties which enables the probing to be performed deeper into the substrate with the thermal wave diffusion. IRR, on the other hand, is able to detect in greater detail the restoration performed in correspondence to what appear as vertical cracks in the paint associated with different local reflectivity properties in the MWIR range in which local restoration of the paint layers has been performed. Such a feature is only partially detectable by IRT.

Finally, it is worth remarking that, unlike IRR, IRT can depth-resolve the position of the detected features. In this respect, the development of IRT signal models for a quantitative analysis of the experimental data originating from optically semi-transparent samples containing graphic features buried within their volume is currently underway. This involves the development of both numerical routines for solving the model equations and experimental procedures for the evaluation of the sample optical and thermal properties on which the IRT signal depends.^44,45

IRT has also been proven to be an effective tool for the investigation of frescos realized on a smooth coat of wet plaster (intonaco) applied on a wall. This enables the applied water based paints to diffuse into the intonaco and, with the setting of the plaster, they become a permanent part of the wall.

Fresco paintings can also undergo permanent damage because of different reasons, with subflorescence and efflorescence phenomena being the main ones. Both phenomena involve water migration through the wall and subsequent evaporation that leads to the formation of crystal deposits which can induce local detachments in the plaster layer. The traditional “finger tapping” technique that is still often used by restorers to detect these detachments is subjective and cannot provide any quantitative assessment. IRT has also turned out to be a reliable tool for the detection of defects buried within the frescos.^46–50 Given the larger thickness of the plaster layer compared with that typical of panel paintings, the step-heating scheme has been adopted for the purpose. Such a scheme makes use of a low-intensity heat source, which allows a long heating time to probe the sample over a large depth without causing any damage. Particular procedures are adopted to correctly distinguish the defects associated contrast from that due to the painted overlayer. To achieve this, suitable post-processing procedures of the IRT data are sought, and also the use of heating procedures different from the conventional halogen lamp light absorption is envisaged. For example, hot air fan⁴⁸ or ceramic lamps emitting in the range between 2 μm and 10 μm have been successfully used.⁵¹ The latter source takes advantage of the uniform optical properties that most of the mural painting colors show in that spectral range. As already found in the case of panel paintings, special attention has also been given to the realization of robotized systems,⁵² allowing to automatically scan the IR camera in order to acquire the IRT images over large historical frescos. In such investigations, the IRT transmission scheme has been adopted to inspect the frescos in a medieval chapel where the heating source was the solar radiation illuminating the outer surface of the walls. Solar irradiation can be regarded as a periodic stimulus and, consequently, the recorded IR image sequence was processed according to the lock-in technique. In Fig. 4, the amplitude (left) and the phase (right) images obtained by automatically scanning a wall of the chapel are shown. Under the assumption of both a uniform solar irradiation and a constant wall thickness, the contrast features shown by the recorded images have been considered to be mainly originating from local variations in the thermal properties and, more specifically, from the local humidity content of the wall. In fact, the thermal diffusivity of wet plaster turns out to be lower than that of the dry one, such a difference becoming more relevant with increasing the water content. Consequently, thermal waves propagating through materials with different humidity content are expected to undergo different attenuations as well as relative phase shifts.

Sometimes ancient frescos can be found lying beneath a lime layer. Research is underway to apply IRT to study such frescos. Such an activity involves both numerical modeling of the IRT signal originating from buried graphical features⁵³ and the development of innovative IRT experimental approaches. Given the partial transparency that lime layers show in the spectral range corresponding to wavelengths larger than 40 μm, in Ref. 54, the use of a blackbody source at 500 °C as the heating source has been explored. Under such circumstances, the heating IR radiation is partially transmitted through a thin lime layer and eventually absorbed by the buried pigments in which it can induce a local temperature rise, thus leading to an overall improvement of the visibility of the buried graphical features. As seen for the panel paintings, integrated approaches combining the use of IRT and other techniques are being considered. For instance, step-heating IRT has recently been adopted⁵⁵ to assess the structural integrity of ceilings possibly damaged by recent earthquakes. In this study, a stream of hot air has been used to induce the sample heating to avoid damage to the decorative apparatus. In the same study, fiber-optics diffuse reflectance spectroscopy (FORS) measurements have also been carried out in the VIS-NIR spectral range. Such measurements were aimed at characterizing the pigments through a comparison of their diffuse reflectance spectra with the ones available on databases.

In this kind of artwork, both VIS light absorption and IR emission take place at the sample surface so that the IRT signal is proportional to the temperature variation occurring at the sample surface. Thus, the IRT signal time dependence is not significantly affected by the sample optical properties but only on the structural and thermal properties which, in some cases, can be known with sufficient accuracy to enable the assessment of quantitative characterization of the investigated items. With this aim, research activity is currently focused on the development of suitable numerical algorithms for the interpretation of the retrieved IRT signal. The validation of such tools can then be obtained by applying them to the analysis of the results obtained in test samples reproducing most of the subsurface features typically present in bronze statues.

One of the main tasks of the IRT investigations carried out on bronze statues is to detect and characterize the workings undertaken after the main casting is done, and the resultant information is extremely valuable for both scholars and restorers. According to the procedure followed in the manufacture of ancient bronze statues, several kinds of workings were carried out after the casting^56–58 for artistic purposes or for the repair of defects, such as casting faults and openings, like those made to insert the internal armature. All these workings were then masked by a finer tooling and polishing of the surface to obtain the final appearence of the statues. However, such processes caused local inhomogeneities in the structures of the statues which can then be detected by means of IRT.^59–61 Therefore, in the last few years, experimental studies on bronze statues have also begun to test the ability of IRT to detect such features. With this aim, thermographic measurements have been successfully carried out in both test samples and manufactured copies of original bronze statues in which different kinds of mechanical repairs were applied and then completely concealed by the surface patination.^59,60 In these investigations, the PT measurements performed with a large image recording rate have been proven to be an effective tool to detect even shallow features. Such a rate is necessary to selectively probe the features located at different depths due to the short duration time (<1 s) of the transient signal resulting from the good thermal transport properties of the bronze alloys. On the basis of such preliminary studies, over the last few years, restorers have started applying PT to inspect original artwork, in which, among others, it has been proven to constitute an effective tool to distinguish between the different methods used to mend the openings, the main ones being the so-called mechanical and metallurgical repairs.^62,63 While a mechanical repair consists of regular-shaped patches mechanically forced in pre-applied shallow cavities, metallurgic repairs are cast by pouring molten bronze into irregularly shaped cavities. Thanks to the larger temperature variation at the patch edges because of the presence of a thermal barrier at the interface with the statue main body and also of the smaller thickness of the metal at such positions,^64–66 the recorded thermograms allow prompt visualization of the patch shape and, hence, the identification of the applied repair. As an example, the thermograms recorded on the Capitoline She Wolf bronze statue in the areas evidenced in Fig. 5(a) clearly indicate the presence of a mechanical [see Fig. 5(b)] and of a metallurgical [see Fig. 5(c)] repair, respectively. In this respect, it is worth pointing out the importance of distinguishing between the two different kinds of repairs. While the irregularly shaped metallurgical repairs can be assumed to be due to some unwanted faults that would have occurred during the main casting of the statue, the mechanical ones can be considered to correspond to apertures that had been specifically planned for the removal of the elements of the armature used to support the statue during the casting. Based on such an idea, a model has been proposed for the structure of the armature adopted in the manufacture of the She Wolf, according to which, the observed mechanical repairs shown in Fig. 5(a) were applied to mend the entry portholes of two of the bars that constituted the armature.

PT has, therefore, proved to constitute a valuable tool to investigate the workings carried out on statues after the main casting is done. In this respect, the use of PT as a diagnostic tool will, in the future, benefit from the recent introduction of high-resolution IR cameras enabling a full-frame image recording in a frame rate range extending up to 1 kHz. Such an opportunity would allow a more accurate characterization of the subsurface features and, more specifically, of the shallow ones, thanks to the increased time resolution.

An approach similar to the ones mentioned above is currently under development for the detection and characterization of applied insertions. Those applied in the external parts of the main body of the statues are not always motivated by the need to repair elements. In fact, additional parts can be inserted because of artistic reasons such as the need to obtain features with different colors with respect to the main statue.⁶⁷ As already seen in the case of repairs, the application of such parts can be accomplished by using either mechanical or metallurgical procedures. To distinguish between the two, which is of key importance for the study of the statue manufacture, a new approach has recently been developed based on the monitoring of the heat diffusion rate taking place at the interface between the inserted elements and the main body of the statue.⁶⁸ In the case of a metallurgical insertion, the cooling of the molten part in contact with the main body of the statue is expected to give rise to a good thermal contact at the interface. In contrast, in the case of mechanical insertions, the thermal contact is expected to be less effective due to the presence of possible residual air gaps. The consistency of such hypothesis has been validated through measurements carried out in the test samples of known thermal and geometrical characteristics simulating either a metallurgical or a mechanical insertion. With this aim, a numerical model simulating the IRT signal originating from these kinds of samples has also been developed. Following the successful application of this approach in test samples,^68,69 the procedure was subsequently also applied to the characterization of insertions in original artwork such as the Boxer at Rest and the Hellenistic prince hosted in the Museo Nazionale Romano in Rome.⁶⁸  Figure 6(a) shows the thermogram obtained on the head of the Boxer, while Figs. 6(b) and 6(c) show the time dependence of the PT signal obtained on the lips and on the swelling under the right eye, respectively. For the former, after the initial $1/ t$ time dependence as in the thermally homogeneous samples, the signal shows only a slight slope change related to the different thermal properties of the lip material (copper) and the substrate (bronze) but with a good thermal contact between the two, indicating a likely insertion of the lips in the molten bronze substrate. In contrast, for the swelling, in the PT signal decay shown, after an initial part corresponding to the heat diffusion through the swelling homogeneous layer, a substantial slowing down occurs, indicating the presence of a thermal barrier at the interface with the bronze because of a likely mechanical insertion of the swelling. In both cases, the continuous curves correspond to the signal calculated according to the developed numerical model.⁶⁸ 

In order to carry out the quantitative characterization of the detected features, it is necessary to evaluate the thermal diffusivity of the bronze alloys that are most commonly used for the manufacture of the statues. Experiments in this direction are currently being performed in test samples and in parts of original bronze statues. For instance, in Ref. 61, the flash method has been used to measure the thermal diffusivity in different parts of a bronze statue in which both sides could be accessed by IRT instrumentation. Thanks to the obtained diffusivity values, it was possible to evaluate the thickness of the bronze layer in various parts of the statue, the depth of some casting faults, and the depth profile of some patches/insertions. The thermal diffusivity of the characterized bronze statues has been shown to be dependent on the alloy composition, especially the ones containing mainly copper and tin which, in ancient times, were the primary materials of the bronze used for the manufacture of statues. In this respect, in the next future, it will be of primary importance to accurately evaluate the correlation between the alloy composition and the thermal diffusivity value so that the knowledge of D could allow the retrieval of information about the composition of the bronze alloy. Conversely, it must be considered that the local measurement of D may result very difficult in the statue parts corresponding to irregular shaped surfaces. In these cases, the value of D could be estimated once the alloy composition is established, for instance, by means of x-ray fluorescence.

For a more effective representation of the IRT results concerning the detection of the subsurface features, in particular, an integrated methodology combining PT with 3D laser scanning technologies has been recently proposed to obtain a 3D thermographic representation of the bronze subsurface features.⁵⁹ The procedure consists of the simultaneous acquisition of both sample geometrical data by the 3D scanning device and thermograms by the PT setup. Such a procedure provides the possibility of translating and rotating the position of the features revealed by the IRT into the geometrical space defined by the 3D scanner.^59,65 Moreover, it has also been possible to obtain different 3D-IRT reconstructions corresponding to different delays with respect to the heating pulse and, hence, to different depths into the investigated bronze statue like the case of the Capitoline Brutus whose 3D-IRT reconstruction is shown in Fig. 7. Such a reconstruction constitutes a useful tool for different kinds of end users, enabling to effectively show surface and subsurface features as revealed by PT. Moreover, this 3D representation can, in perspective, be used to also include the results obtained by other non-destructive imaging techniques such as the chemical composition maps obtained by spectroscopic techniques.

In the last few years, the application of IRT to the inspection of library materials and, more specifically, of books⁷¹ has attracted considerable attention from researchers. As shown in the following, several research activities are underway in order to improve the effectiveness of the analysis carried out by means of IRT to the main parts of books, namely, the binding, the writing support, and the text and/or decorative elements.⁷² 

As regards binding, the PT technique can provide information on both the structure and the inherent damage⁷³ existing in the binding elements without dismantling the books. Among others, PT has been used to evaluate the adhesion state of the different parts of binding or to assess the extension of woodworm attacks on the wooden covers of books.⁷⁴ In this respect, research activities have also been focused on the development and application of algorithms for the processing of the IRT raw data aimed at enhancing the defect visibility. For instance, the wavelet transform thermography (WTT)⁷⁵ and the higher-order statistics thermography (HOST)⁷⁶ processing techniques have been used in Ref. 77 to analyze the IRT images obtained on the binding of an ancient book. In addition, multidisciplinary experimental approaches combining the use of IRT and other complementary techniques have also been initiated in this field. In Ref. 77, IRT, IRR, and optical coherent techniques such as digital speckle correlation and holographic interferometry have been simultaneously applied to the inspection of bookbinding, such techniques being able to probe the presence of defects over different depth ranges into the inspected item. Such an approach also makes use of nuclear magnetic resonance measurements to locally evaluate the hydration degree of a cellulose amorphous structure which, if too large, may cause significant damage to the bookbinding structure.

Recently, IRT investigations of bookbinding have led to the development of another research activity. Such an activity concerns the detection and the analysis of the scraps of old books and manuscripts used for bookbinding and that lie between the end papers and the covers of books.⁷⁸ In particular, IRT has been demonstrated to be the only available tool for revealing the text written on these buried scraps, which may be very interesting from both a historical and an artistic point of view. For instance, Fig. 8(a) shows the picture of the back endpaper of an 18th-century book. The recorded thermograms [Fig. 8(b)] show the presence of two superimposed written parts belonging to different fragments. In particular, the one indicated by the red frame and arrow [Fig. 8(c)] corresponds to a paper fragment directly attached to the back surface of the endleaf so that it becomes clearly visible right after the application of the heating pulse. In contrast, the text indicated by the blue frame and arrow [Fig. 8(d)] becomes clearly visible in the thermogram recorded 300 ms after the pulsed heating due to the larger depth of the paper fragment on which it is written. In this respect, it is worth noting that the detection of such hidden texts in the IRT images is ensured, thanks to the following two mechanisms.

The first one is relevant in the case of subsurface features located at a depth range smaller than the penetration depth of the heating VIS light as in the case of Fig. 8(c) Here, the different absorption properties in the visible range of the written parts, with respect to the surroundings, result in a different local temperature rise and, hence, in a contrasted IR emission map in the recorded thermograms. On the other hand, when the depth of the subsurface element is large and it cannot be directly heated by the incident light as in Fig. 8(d), the buried text may be reached by the diffusing heat front and the text contrast in the thermogram will be associated with a locally different value of the IR emissivity with respect to the surroundings. This second mechanism makes IRT capable of reading subsurface graphical elements deeper down the artwork volume compared with the reflectographic techniques. In order to improve the effectiveness of such an IRT application, numerical IRT signal models for the description of the IRT signal originating from buried features have been recently proposed, with their development still in progress.⁴⁵ The aim of such models is to evaluate the influence exercised by the values of the involved optical and thermal properties of the paper/ink structure on the detectability of the buried texts with the consequent possibility of using the developed models to post process the data to improve the readability of the writings. The models need to take into account effects like lateral heat diffusion and scattering of the IR radiation in the overlaying paper leaves. The developed models could also be used to evaluate the depth of the detected buried text. In this respect, IRT experimental configurations for the evaluation of the text substrate opto-thermal properties have been recently reported.⁴⁵ 

Ancient manuscripts may undergo severe degradation due to different reasons like biological attacks or inappropriate ambient conditions of the hosting environment which, among others, may lead to a loss of the readability of text parts due to the deterioration/loss of the ink solid component. In this respect, IRT has been fruitfully applied to recover the otherwise unreadable text.⁶⁵ Such a possibility is granted by the presence, in the text support, of the ink binding component which may absorb the UV–VIS component of the incident heating light, and also by the different IR emission properties of such binding components with respect to those of the surrounding material, both inducing the text contrast in the recorded thermograms. Further development of this kind of IRT application should focus on the identification of the most suitable spectral ranges of both heating light and the IR detection range, which may lead to an enhancement of the text contrast in the recorded thermograms.

Multidisciplinary approaches are now being considered where IRT is used in combination with other complementary techniques for an in-depth inspection of the decorative apparatus of texts,⁷⁹ and, in particular, of illuminations. In Ref. 80, both IRT and IRR have applied to the identification of underdrawings and pentimenti lying beneath the pigment layers of illuminations. In addition, the gilding conservation state and the adhesion to the parchment support have been also investigated by means of IRT, while their composition has been analyzed by both XRF and energy dispersive spectroscopy techniques.

In the previous paragraphs, a large variety of applications employing thermography for the study of cultural heritage items were reviewed, highlighting the research and development trends aimed at implementing the diagnostic capability of this technique and reducing the impact of the measurement technique on the artifact to comply with the requirements of conservation policies. From this review, some trends emerge that are common to many of the research studies carried out on the different investigated items. Among them, the following stand out: the integration of thermographic imaging with that provided by other non-destructive analysis methods, the realization of a multilevel three-dimensional thermography, and the development of specific data processing procedures to improve the readability of the features in thermograms and of models for IRT signal generation, allowing a quantitative analysis of the elements identified in the thermograms.

Regarding the combined use of thermography with other imaging methods, of particular relevance are those, such as the reflectography in SWIR and MWIR ranges, capable of making use of the same detector. Their combined use generally allows the characterization, in a complementary way, of subsurface elements, as in the study of paintings in which, while the reflectograms can highlight details of the preparatory drawing and of restorations and repaintings, the thermograms can also reveal some characteristics of the substrate. In other applications, such as in the study of inks in books and documents, it has been seen how the contrast variations in the reflectographic images obtained in different spectral ranges of the IR spectrum can provide information on the composition of the ink used for the writing, while the thermographic images, obtained by the same detector, guarantee the readability of buried or faded texts. Regarding the possibility of combining the use of IRR and IRT, scenarios are opened in which thermographic and multispectral reflectography investigations can be carried out with the same camera and the use of filters in different ranges, over a relatively extended IR spectrum (1.5–5 μm). The use of the same camera would enable, for example, when illuminating with a multispectral step like light source, the recording of reflectographic images at short times and thermographic ones after the sample heating becomes effective.

When using different detectors (for example, InSb and Si), this approach allows performing multispectral imaging by integrating digital images obtained throughout a significantly extended spectral range, from UV to MWIR. The combined use of thermography and other imaging techniques in such an extended spectral range also gives the possibility of using the same light excitation process to simultaneously obtain images by different detection techniques. For example, by means of UV light illumination, UV reflectography images, fluorescence images in the VIS range, and thermographic images in the IR range could be obtained. Such an approach may allow a meaningful comparison of the images of surface and subsurface elements of the artifact in applications such as the study of inorganic and biological patinas or the analysis of pictorial layers.

Another interesting perspective concerns the combined use of IRT with other imaging techniques to obtain a 3D optical image representation of the investigated items. Such representation would allow a visualization of the structural and the subsurface elements of the artifact in their geometrical context and, consequently, an improvement of the effectiveness of the display of the results. In addition, by combining thermograms corresponding to different values of either the delay time or the modulation frequency, it is possible to obtain 3D renderings of subsurface features located at different depths into the sample volume. So far, the procedure adopted for the 3D image reconstruction of the recorded thermograms consists mainly of the simultaneous use of an IRT apparatus and of a 3D scanner, the latter adopting either structured light or a laser scanning imaging system to reconstruct the shape of the investigated object. In fact, by knowing the relative position of the IR camera and of the 3D scanner, the thermographic features can be matched to the spatial coordinates obtained by the 3D scanning technique. However, the improvements shown by a recently upgraded IR camera such as the large spatial resolution (1.3 × 10⁶ pixels) combined with the large spectral range (1.5–5 μm) opens the possibility of obtaining 3D image representation by adopting 3D scanners in the SWIR or MWIR range. It is worth remarking that such an approach may make use of only the IR camera to record both the thermograms and the shape of the investigated item, thus significantly simplifying the overall procedure.

Regardless of the specific artwork category, research efforts need to focus on the design of algorithms for the postprocessing of IRT data to improve the visibility of the detected subsurface features and on the proposal of new modulation schemes of heating beams allowing an in-depth inspection of artwork while limiting the effects of heating. Finally, the development of models for the IRT signal, which will be useful for the analysis of the experimental data, will play a crucial role in further improving the effectiveness of the application of IRT and, in general, of photothermal techniques to the inspection of cultural heritage items.

In this work, we have reported on the results that have been obtained in the field of the application of photothermal techniques to the study of cultural heritage. It was shown that photothermal techniques can be applied for a nondestructive analysis of several kinds of cultural heritage items, thanks to their ability to detect subsurface features in objects and to depth-resolve their position within the object volume. The application of infrared thermography, in particular, to the study of items such as paintings, bronze statuary, and books has been overviewed. The advantages and drawbacks of infrared thermography with respect to other techniques commonly used for these kinds of applications have been discussed. Finally, the future perspectives of the application of infrared thermography to this field are illustrated by pointing out further developments that can be obtained to improve the effectiveness of its application to the study of cultural heritage. These include aspects like the integration of thermographic imaging with that provided by other non-destructive analysis methods, the realization of a multilevel three-dimensional thermography, and the development of specific data processing procedures and of numerical models to improve the readability of the features in thermograms and to enable a quantitative analysis of the elements identified in the thermograms.

The authors are very gratefully to Padre Chihade Abboud, Rector of Santa Maria in Cosmedin Basilica for providing the possibility to perform the investigations on the altarpiece.

The data that support the findings of this study are available within the article.