spectral characterization of a digital still camera’s nir … › ~pfsmet › papers › a1_33...

3456 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 47, NO. 10, OCTOBER 2009

Spectral Characterization of a Digital StillCamera’s NIR Modification to Enhance

Archaeological ObservationGeert J. Verhoeven, Philippe F. Smet, Dirk Poelman, and Frank Vermeulen

Abstract—Scholars using still cameras to take (mostly) obliqueimagery from a low-flying aircraft of various possible archaeolog-ically related anomalies can be defined as aerial archaeologists.At present, as well as in the past, aerial/air archaeology has beenacquiring data almost exclusively in the visible range of the elec-tromagnetic spectrum. This phenomenon can largely be attributedto the critical imaging process and sometimes unconvincing re-sults related to the film-based approach of near-infrared (NIR)photography. To overcome the constraints of detecting and inter-preting only the varying visible colors in vegetation (the so-calledcrop marks), while still maintaining the flexible and low-costapproach characteristic for aerial archaeology, a consumer digitalstill camera was modified to capture NIR radiation. By its spectralcharacterization, more insight was gained into its imaging prop-erties and necessary guidelines for data processing, and futureimprovements could be formulated, all in an attempt to bettercapture the archaeologically induced anomalous growth stresses incrops.

Index Terms—Aerial archaeology, camera characterization,crop mark, digital photography, near-infrared (NIR).

I. INTRODUCTION

A. Aerial Archaeology

THE TERM “aerial archaeology” encompasses the entireprocess from the acquisition and inventory of imagery to

the mapping and the final interpretation. It comprises the wholestudy of all sorts of archaeological remains by using informa-tion acquired from a certain altitude: digital or film-based low-altitude aerial photographs, satellite imagery, lidar, radar, etc.The majority of source data used by most aerial archaeologistsare acquired from the cabin of a low-flying airplane usingsmall- or medium-format handheld cameras with (generally)uncalibrated lenses, mostly capturing oblique imagery. Al-though this specific type of data acquisition may seem strangeto the nonarchaeological community, the noninvasive approacheasily yields interpretable imagery with abundant spatial detail,

Manuscript received November 20, 2008; revised February 10, 2009. Firstpublished August 7, 2009; current version published September 29, 2009. Thiswork was supported by the Fund for Scientific Research—Flanders (FWO).

G. J. Verhoeven and F. Vermeulen are with the Department of Archaeol-ogy and Ancient History of Europe, Ghent University, 9000 Ghent, Belgium(e-mail: [email protected]; [email protected]).

P. F. Smet and D. Poelman are with LumiLab, Department of Solid StateSciences, Ghent University, 9000 Ghent, Belgium (e-mail: [email protected]; [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TGRS.2009.2021431

is extremely flexible, might be cost efficient (certainly whencompared to other prospecting methods and applied in previ-ously unexplored areas), and is driven by the specific nature ofthe archaeological anomalies.

Archaeological remains such as settlements, graveyards, androads can show up on the surface in a number of ways. Asidefrom still-standing material relics (e.g., churches, bridges, andfortifications) and partly eroded structures (e.g., earthen banks,mounds, and ditches), most of the features that can be viewedfrom above are the remains of buried archaeological sites.Whereas the first type of archaeological features is directlyvisible, the second type—often referred to as earthworks—ismostly recorded from the air when thrown into relief by low-slanting sunlight (sometimes referred to as shadow marks) andin northern Europe by differential snow accumulations or dif-ferential melting of snow or frost. The buried or leveled remainsmight be disclosed by distinct tonal differences in the (usuallyploughed) soil (soil marks) or differences in color and/or heightof vegetation on top of the remains (crop/plant marks), withthe variations in the subsoil being the prime movers in theircreation. In other words, archaeological residues must exhibita certain localized contrast in their surrounding matrix to bedetected [1]. Although these marks are mostly discovered,photographed, and mapped using visible light, this paper willexplore how these anomalies, particularly crop marks, canbenefit from detection and interpretation by low-cost digitalaerial imaging of near-infrared (NIR) radiation. Consequently,the nature of crop marks needs to be considered first.

B. Crop Marks and Related Plant Reflectance

Subsurface archaeological remains such as pits or trencheswill often be filled with organic material and/or new soil, whichhas greater moisture retention than the surrounding matrix. Inperiods of drought, these soils might have a favorable effect onthe crops, allowing the plants to grow luxuriantly and for anextended period of time. The adjacent plants will be less talland thinner and ripen quicker, leading to differences in chromaand/or plant size that can be seen from above as positive cropmarks [Fig. 1(a)].

In unfavorable situations [e.g., plants growing over buriedstone walls or floors—Fig. 1(b)], weaker and shorter plantsmight occur, in which case negative crop marks are yielded[2]–[9]. Speaking in more technical terms, such adverse sit-uations put a certain stress on the vegetation, hence blockingthe growth, development, or metabolism of the plant. It is the

0196-2892/$26.00 © 2009 IEEE

Authorized licensed use limited to: University of Gent. Downloaded on September 25, 2009 at 03:15 from IEEE Xplore. Restrictions apply.

VERHOEVEN et al.: SPECTRAL CHARACTERIZATION OF A DIGITAL STILL CAMERA’S NIR MODIFICATION 3457

Fig. 1. (a) Positive and (b) negative crop marks (adapted from [2, Fig. 13]).

Fig. 2. Kodak Ektachrome Professional Infrared image of a dense archaeo-logical landscape containing Neolithic and Roman features [29, Fig. 8].

stress-related loss of chlorophyll—a green pigment that can befound in all green plants and largely absorbs incident visiblewavelengths in the blue waveband (centered around 450 nm)and red (around 650 nm) spectral region [10]–[12]—whichinduces an increased visible reflectance in the green–yellow–orange waveband and the red chlorophyll absorption regionaround 670 nm [13], [14]. Consequently, the plant’s domi-nant green color disappears in favor of a yellowing discol-oration, which is a phenomenon called chlorosis [15]–[17]. Byrecording the reflected portion of the visible radiation, aerialphotographs thus allow the remote assessment of vegetationstatus [18].

However, aerial archaeologists have sometimes acquired im-agery using other parts of the electromagnetic (EM) spectrum(Fig. 2), particularly the NIR waveband (see [19] for an ex-tensive overview). In the NIR (700/750 to 1400 nm), pigmentabsorption is extremely low [20], and the leaf’s internal cellularstructure (more particularly the structure of the spongy meso-phyll) effects a very high and diffuse reflectance [12], [21]–[23].In the case of diseased, senescent, and heavily nutrient-deficientvegetation, reflectance can significantly drop in the photo-graphic NIR region [24]–[28], with an absolute change in theNIR reflectance that might be far more noticeable than the re-flectance increase in the visible band (for an in-depth overviewof a plant’s physiological- and morphological-state-relatedspectral differences in the NIR, consider [19]). Although imag-ing reflected NIR has been recognized as potentially beneficial,a film-based approach has certain inherent drawbacks (e.g.,the requirement for cooled storage and transportation of emul-sions, inappropriate exposure determination, narrow exposurelatitude, and relatively weak sensitivity), making the completeNIR image acquisition and processing workflow costly andcomplicated, with a final outcome that is rather unpredictable.

C. Digital NIR Acquisition

Since the advent of digital photographic cameras [also calleddigital still cameras (DSCs)], the acquisition of such NIRimagery has enormously been simplified, because their siliconimage sensors are very sensitive to this invisible radiation, witha so-called cutoff wavelength λc at circa 1100 nm [30]–[32]. Inaddition to the digital image sensor, the whole imaging array ofmost one-shot DSCs also consists of a microlens array, whichis used to increase the amount of photons impinging on thesensor’s photodiode (i.e., the light-sensitive area that collectsphotons, hence creating one pixel of the final digital image), anda color filter array (CFA), which is a mosaic pattern of coloredfilters positioned above the photodiodes [Fig. 3(a)] [31], [33]–[35]. As every photodiode of the image sensor has such a filter,only a specific spectral range can be transmitted, subsequentlygenerating a charge in the photodiode [Fig. 3(b)].

Although both the sensor technology and the arrays of mi-crolenses and colored filters are responsible for some variationin the spectral responses of DSCs, it is safe to state that mostimaging matrices are very responsive to NIR radiation (for amore in-depth discussion, consider [36]). To cut out the image-degrading effect of these nonvisible wavelengths, camera man-ufacturers place an NIR-blocking filter in front of the sensor[37]–[39]. By removing this optical element and replacing itwith a visibly opaque filter, all visible wavelengths are removedbefore they reach the sensor, allowing only NIR photons topass. Such a modification hugely increases the DSC’s sensitiv-ity to NIR, while retaining the facility to view through the lens(impossible in the film-based approach of pure NIR imaging).

Using a dedicated NIR DSC also deals with most of thedifficulties presented by film. Additionally, digital solutionsoffer enhanced quantum efficiencies (QEs) and larger dynamicranges [41], [42] when compared to analog approaches, whichmeans that the former can be applied in far-from-optimal oper-ational conditions.

Moreover, a DSC’s linear response to radiation, as well asits direct feedback on accurate focusing and exposure, enablesa very consistent output. Finally, DSCs are suited for mappingpurposes, as they do not suffer from geometric film distortions[43], [44]. In spite of these major advantages, the application ofdigital NIR imaging with DSCs was never really investigated inarchaeological reconnaissance.

Using imagery generated by such a modified DSC and con-ventional frames from a simultaneously operated unmodifiedDSC, Verhoeven [19] gives an overview of situations in whichthese easy-to-use NIR-imaging instruments might be archae-ologically advantageous. Specifically, by comparing both datasources, the author demonstrated the potential of this approachto overcome the constraints of detecting and interpreting onlythe varying visible colors in vegetation, while still maintaininga flexible and economic approach (in terms of imaging instru-ments).

This paper further explores the possibilities of such con-verted DSCs in extracting even more meaningful informationfrom an acquired NIR frame, reporting on the evaluation andquantification (as with any scientific measuring tool) of theintrinsic properties of an NIR modified digital single-lens reflex



Fig. 3. (a) Bayer CFA [36, Fig. 7]. (b) Wavelength versus absolute QE for the Kodak KAF-8300 (adapted from [40, Fig. 5]).

(D-SLR) camera. This assessment of the channel-dependentspectral responses and the accuracy of capturing NIR photonsmight offer significant possibilities in the data processing, inter-pretation, and quantification of the acquired imagery. Instead ofonly using the imagery straight out of the camera, exploiting theDSC’s individual spectral responses should ideally permit thecapture of (archaeologically) induced growth stresses in cropseven better (i.e., enhance the contrast between the archaeologi-cal residue and the landscape matrix [1]).

II. DSC CHARACTERIZATION: MATERIALS AND METHOD

A. Hardware

For the reasons discussed in [45], a Nikon D50 D-SLR wasemployed. The NIR modification of the DSC (hereafter calledD50NIR) was executed by Chen [46], who placed a sort of coldmirror in front of the sensor to block most visible radiation. Thesensor itself, a Sony ICX413AQ APS-C format sensor (calledDX format by Nikon) of the charge-coupled device (CCD) type,measures 23.7 mm × 15.6 mm and contains 3008 effectivephotodiodes in width by 2000 photodiodes in height [47], [48].Above this sensor, an on-chip three-color red–green–blue(RGB) CFA is fitted, with the filters arranged in a Bayer pattern,as shown in Fig. 3(a). Bayer’s pattern features twice as manygreen filters as blue or red filters to improve the sampling ofthe luminance information [49], generating digital imagery withhigher perceived sharpness [49], [50].

As the majority of optical glasses and polymers freely trans-mit NIR [39], most lenses can be used for NIR imaging [37],[51], [52]. On the D50NIR, the Nikkor 20-mm f/3.5 AI-S andthe AF-S DX Zoom-Nikkor 17–55-mm f/2.8 G IF-ED areused for Helikite aerial photography (i.e., remotely controlledphotography by means of a Helikite, a helium balloon with kitewings [53], [54]) and photography from an airplane, respec-tively. Whereas the latter lens is slightly more prone to hot spots(i.e., a brighter area in the center of the image produced by in-ternal reflections) than the fixed-focal-length lens, it allows forzooming, which is often necessary when flying. The prime lensis, however, a top-class performer in the NIR, capable of pro-ducing very crisp and extremely sharp images [55]. Moreover,it features an NIR focus mark. This lens was also used in the

subsequently described spectral analyses. To verify the consis-tency of the results, all tests were repeated with an AF Nikkor50-mm f/1.8 D.

B. Image Acquisition

To identify the NIR behavior of the D50NIR’s completeimaging system (lens + cold mirror + microlenses + CFA +CCD), spectral response data are very important as they rep-resent the digital output of the image sensor per incident lightenergy of a certain wavelength. In the procedure followed, a2800-K tungsten lamp was used as a reference EM source withknown spectral output. A small part of the emission spectrumwas selected with a Zeiss quartz prism monochromator (typeCarl Zeiss M4 QII) in the wavelength range from 400 to1100 nm. Using quartz prisms for wavelength selection is bene-ficial as no second-order contributions, which are typical whenusing a diffraction grating, exist. Nevertheless, it was verifiedthat no spurious light in other than the selected wavelengthrange was present.

Subsequently, a small entrance slit was fitted on the mono-chromator to obtain a Gaussian-distributed narrow-band stim-ulus. The transmitted waveband was then characterized with acalibrated Ocean Optics QE65000 spectrometer (with a wave-length resolution of 0.8 nm) to accurately determine the peakwavelength and the bandwidth, which typically had a full widthat half maximum (FWHM) of 2.8 nm at 600 nm and 5.2 nm at950 nm. Finally, characterization with the spectrometer allowedthe number of photons that passed each selected wavelength tobe determined. The D50NIR was irradiated with its sensor per-pendicular to the output of the monochromator to minimize asmuch as possible the angular dependence of the image sensor[56]. Pictures of the transmitted radiation were acquired atmonochrome EM levels every 5 nm to obtain sufficient datapoints. The D-SLR used a lens aperture of f/5.6 and a total ex-posure time short enough (0.25 s for visual and 5 s for NIR)to make sure that no photodiode became saturated, while theintegration time was still long enough to generate sufficientlyhigh digital numbers [DNs, also called analog-to-digital units(ADUs)], essential for an acceptable signal-to-noise ratio(SNR) (or S/N) and related measurement accuracy. For all im-ages, the D50NIR’s default ISO 200 setting was used, yielding aminimal gain g of 6.57e−/DN with the 12-bit analog-to-digital



converter (ADC). This value was calculated according to themethod described by Berry and Burnell [57] and indicates thenumber of electrons that will cause the DN to increase byone [33], hence corresponding to a linear scaling factor K of0.152 DN/e− (K =1/g).

As it is very important to work with the initially generatedinteger values, using RAW imagery is crucial. In essence, aRAW file is nothing but an array of DNs, each of them gen-erated by one photodiode and proportional to the EM radiationof a certain wavelength range (determined by the colored filteron top) plus some offset due to dark current and bias. Becausethe D50NIR utilizes a 12-bit ADC, the DNs can vary from 0 to4095, corresponding to a tonal range of 212 gradations. Usinga RAW workflow ensures that the imagery for analysis is the“pristine” sensor data, as these files (which can be created bymost consumer and all professional DSCs) were not subjectedto any color-processing algorithms (i.e., white balancing, demo-saicking, tonal curve) by the DSC’s firmware, unlike in-camera-generated JPEGs and TIFFs (for a discussion on the necessityof using RAW in scientific imaging, consider [58]).

C. Image Calibration

Subsequently, the RAW images (called NEF by Nikon,which means Nikon Electronic Format) were imported to TheMathWorks’ MATLAB to measure the DSC’s response to thenarrow-band illuminations but not before calibrating the im-agery by removing some unwanted signals.

In scientific digital imaging, only the stream of photonsthat reach the sensor (i.e., the photon signal) is of interest.However, the light frame captured by an image sensor alwaysencompasses three particular signals: the photon signal, thedark-current signal, and the bias signal/direct-current offset[57]. Unlike the photon signal, which is generated by theaccumulated EM radiation during the exposure, dark currentis a signal that is produced even when the sensor is not illu-minated, due to thermally induced electrons. This dark chargeaccumulates with integration time and is heavily temperatureand ISO dependent. The bias component, which is a small andmostly steady zero voltage offset that occurs even in the totalabsence of illumination, is due to the effects of the electricalcharge applied to the detector prior to exposure [57], [59].

Each of these nonrandom signals has some correspondingrandom variation (i.e., noise) embedded, all three varying ac-cording to the imaging technology used [60]. In addition tophoton/shot noise (σ) and dark-current noise (σd), caused bythe inherently random process of photon arrival and both obey-ing the law of Poissonian statistics [33], [61], there is the signal-independent read/readout/bias noise (σron): the sum of the resetnoise (σreset), the on- and off-chip amplifier noise (σamp-on andσamp-off ), and the quantization noise (σADC) [34], [62]. In theD50NIR, this minimal noise floor was measured to be about1.04 ADU (12 bits) or about 6.83 root-mean-square electrons(i.e., 1.04 ∗ g), an extremely low value that makes the D50NIR

completely photon noise limited when imaging normal signallevels and set to ISO 200.

Hence, the DNs making up an NEF picture are the sum ofthe photon signal (with its corresponding Poisson noise), an

TABLE IAPPROPRIATE SYMBOLS AND UNITS OF

ALL MENTIONED DSC QUANTITIES

unwanted dark-current signal (with Poisson noise), and a biasconstant (with readout noise), mathematically written as (1),with the noise equal to (2) ([57], all symbols are defined inTable I)

Sraw =x

g+

xd

g+ b (1)

σraw =1g

√σ2 + σ2

d + σ2ron. (2)

Due to their randomness, the noise components are difficultto correct. However, the dark-current and bias signals can beremoved during calibration. To reveal the dark characteristics ofthe D50NIR, several sets of five NEF images were shot at darkcondition, each set with a different integration time, startingfrom the fastest possible shutter speed (0.00025 s) up to 1 s,while the DSC was in thermal equilibrium at a constant roomtemperature (20 ◦C).

After linearly reading them out (i.e., omitting the nonlineartonal redistribution normally applied by DSCs) and disregard-ing white balance (WB), the RAW frames were converted to16-bit TIFFs (one averaged version per set), and both the meanand the standard deviation of the output values were plottedversus integration time. The results are presented in Fig. 4(a)and show that this D50NIR has significantly low dark-noiselevels at ISO 200.

However, the sudden drop in maximum dark-pixel valuemakes a particular Nikon characteristic apparent. That is,the firmware runs a median filter when the DSC takes anexposure ≥ 1 s, aimed at reducing the effects of hot pixelsduring long exposures yielded by particular photodiodes withabnormally high dark current. The French astronomer Builfound a way around this [63], by turning noise reduction on andshutting down the D-SLR immediately after the exposure hascompleted, thereby aborting the noise reduction job and savingthe pure RAW image directly from the buffer to the memorycard. When applying this method, it is seen in Fig. 4(b) thatthe mean linear dark current is still not even 5e−/diode (i.e.,0.7 DN ∗ 6.57e−/DN) at an exposure of 5 s, which means thatits error contribution is still negligible (apart from a few hotpixels).



Fig. 4. (a) DNs generated by dark current (+ bias signal) versus exposuretime (in seconds) for very short exposures. (b) DNs generated by dark current(+ bias signal) without subsequent median filtering.

After using this method in the data acquisition, a dark framewas subtracted from all RAW images as in (3), with the totalimage noise mathematically expressed by (4)

Simage = Sraw − Sdark =(

x

g+

xd

g+ b

)−

(xd

g+ b

)(3)

σimage =√

σ2raw + σ2

dark. (4)

Expression (4) clearly shows the noise to slightly increaseby dark subtraction. Therefore, rather than generating a singleframe, a high-S/N master dark frame yielded by averaging tenstacked 5-s dark frames (or 0.25-s dark frames) was subtractedfrom the original image to average the random noise. As themaster dark frame also contains the bias component b, thisoperation corrects for both unwanted signals, making the useof a bias frame obsolete [57], [64], [65]. Third, this approachalso accounts for the possible amplifier glow resulting from aresponse of the photodiodes to radiation emitted by the readoutamplifiers every time the detector is read out [59], although thelatter was not visually attested.

In addition to dark subtraction, calibration also involves theremoval of a multiplicative component by flat fielding [57],[61], [64], [65]. This process corrects the image for photo-response nonuniformity (PRNU) by dividing the dark-subtracted light frame with a master flat frame: an average ofseveral dark-current-corrected images taken from a uniform or“flat” field of light, hence recording dust particles on the lensand sensor, optical vignetting, and photodiode nonuniformity,which is the main cause of PRNU [66].

Fig. 5. Relative response versus wavelength of the Nikon D50NIR with aNikkor 20-mm f/3.5 AI-S.

Finally, all calibrated RAW images were analyzed with apurpose-written MATLAB program. Once the spectral andintensity response of both green filter sets were verified to beidentical, a DN for the red, green, and blue sensor responseswas extracted by averaging over a rectangular section of some15 pixels × 100 pixels in the center portion of every image. Theresulting set of three measured intensities allowed plotting thecolor-filter-dependent relationship between the captured wave-length and the ratio of the DN to the intensity of the emitted ra-diant energy. However, accurate measurement of such a spectralsensor response requires the output signal to be linearly propor-tional to the incident light intensity over a large range of inputlevels. Although this is known to be mostly the case [67] andcertainly to be expected for modern DSCs [68], a coefficient ofdetermination, i.e., R2 > 0.99 (calculated for both the completeCFA and all three color channels), confirms the almost perfectlinearity of the photometric response below saturation for thisCCD, an observation that was also reported in [69].

III. DSC CHARACTERIZATION: RESULTS AND PROCESSING

A. Spectral Response Curves

Fig. 5 displays the relative spectral sensitivity response ofthe different photodiodes in the D50NIR to the 2800-K lampas measured with the procedure explained above. The graphdescribes the way in which the whole imaging matrix respondsto particular wavelengths. By repeating the same procedurewith an AF Nikkor 50-mm f/1.8 D, it was verified that theimpact of the photographic lens can be ignored to a large extent.Only from 740 nm onward do the eleven lens elements of theNikkor 20-mm f/3.5 AI-S [70] slightly decrease the NIR trans-mission rate [71] compared to the 50-mm lens (which consistsof only six lens elements [72]). This fact confirms that normalphotographic lenses are highly transparent to NIR radiation,although—strictly speaking—they also have a specific spectralabsorption response.

In addition to transmitting radiation in specific spectral bandsof the visual spectrum, the colored filters thus also functionas wavelength-specific filters in the NIR range, allowing thephotodiodes to capture information in particular spectral bands.



From the curves, it is clearly seen that the spectral sensitivityis almost negligible for visible light with wavelengths below650 nm, corresponding to the cut-on frequency of the NIR-pass filter in front of the CCD. Starting at about 660 nm,the red photodiodes are most sensitive for deep-red to NIRwavelengths, reaching a maximum at 730 nm. Above thisvalue, the QE markedly drops due to generated electrons oftenrecombining before reaching a sensor’s depletion region wherethey are stored [34].

The blue filter locations are, however, totally insensitive forthe entire visible part of the EM spectrum, as their sensitivityonset lies at 780 nm, rapidly increasing to a maximum responseat around 815 nm. The spectral range of 795–875 nm at halfmaximum indicates that most information is gathered beforethe moisture-sensitive NIR trough starting at about 940 nm[73], [74], making the blue-filtered diodes particularly sensitiveto vegetation density or biomass [10], [12], [75]. Because thegeneral spectral response in the blue channel is much weakerthan the green and red responses, it is best to expose with asomewhat longer-than-normal integration time. This will ef-fectively counter high noise levels, as the following equationshows that the SNR increases with the square root of all photonscaptured by the diode [33], [61]:

SNRx =√

x. (5)

Finally, the green diodes show an intermediate spectral be-havior, being responsive to EM radiation from 680 nm onward,until they also reach a maximum at about 815 nm. On thelong-wavelength side (> 820 nm), the similar response ofthe particular diodes indicates that the RGB filters becomenearly completely transparent to the incident radiation, untilthe imaging matrix becomes the perfect equivalent of a mono-chrome detector at around 850 nm, which means that all filteredphotodiodes are equally sensitive to the incoming radiation. Forwavelengths longer than 1000 nm, the D50NIR’s QE becomesextremely low, due to the inherent wavelength-dependent lowabsorption coefficient [34]. On the other side of the spectrum,the sensitivity in the wavelength range from 400 to 650 nmis extremely low, as one would expect from a good visible-blocking filter. Only the green and red photodiodes show avery small response, with green spectrally peaking at 565 nm.Nevertheless, the contribution of these wavelengths to the finaloutput can safely be ignored.

B. New Spectral Bands

NIR imagery generated by the D50NIR has already beenused in archaeological research [19], [36], [45]. However, thespectral characterization described above allows one to gobeyond the initial approaches in which the default output wasused. Because this analysis has clearly revealed the unequalspectral responses of each photodiode type, spectroscopic in-formation can be extracted by differentiating between the red,green, and blue channels. The normalized spectral responseafter subtraction and addition of particular channels is shownin Fig. 6. These mathematical operations make sense, as allthree diode types have the same transmittance on the long-wavelength side, whereas the blue and green spectral responses

Fig. 6. Red channel minus the green channel (R − G), the blue channelsubtracted from the green channel (G − B), and the blue channel added tothe green one (G + B). The peak response of each band is normalized to unity.

TABLE IIALL WORKABLE BANDS GENERATED BY THE D50NIR

completely fit within the response ranges of the green and reddiodes, respectively. This way, the blue pure NIR componentcan effectively be filtered out of the green channel, whereassubtracting the green from the red channel seriously narrows thebandwidth of the latter. Adding the green to the blue band, onthe other hand, creates a new spectral range that peaks at around815 nm, with a better response in the 750–900-nm range, wherea plant’s maximum NIR reflectance lies [76]. Table II gives anoverview of all primary and newly created bands that can beworked with and their close resemblance to particular spectralbands acquired by satellite sensors, although for the purposesof this study, only the archaeological potential of the bandsdisplayed in Fig. 6 is exploited (see Section IV). First, however,one extra elementary processing step is explained.

C. Demosaicking

Apart from the few DSCs that have a Foveon X3 sensor,single-shot DSCs usually feature one CCD, complimentarymetal–oxide–semiconductor, n-channel metal–oxide–semiconductor, or junction field-effect transistor sensorwith an additional CFA to allow one particular spectral band tobe captured by each photodiode. Consequently, a mathematicaloperation must be executed to fill in the DNs for the othertwo bands, which is a process commonly referred to asdemosaicking, color reconstruction, CFA-interpolation, or de-Bayering (in case a Bayer array is used). Given the widespreaduse of CFAs, a large range of linear and nonlinear algorithmshas been created to reconstruct the final RGB image as accu-rately as possible (e.g., [77]–[82]). However, these methods



Fig. 7. Processed images from the same aerial picture taken with the D50NIR. (a) RAW file developed by Capture NX. (b) Same RAW file linearly developedin dcraw. (c) Contrast-enhanced version of (a). (d) Output after a simple mathematical operation (6) on version (b).

were designed to demosaic information from the visibledomain, and the assumptions underlying most of them maynot hold for NIR wavelengths, making them sometimesunsuited for interpolating missing information in NIR imagery.Previous research by Verhoeven [58], however, indicated thatthe adaptive homogeneity-directed demosaicking algorithm[83] performed very well in this invisible domain. As thisalgorithm is implemented in the program dcraw, this softwarehas been used to demosaic all NEF images. Moreover, this freeANSI C RAW decoder works on any operating system and iscapable of writing reconstructed 16-bit TIFF files [84] withoutapplying any tonal/gamma curve or WB (omitting the lattertwo is often of utmost importance in scientific applications[58]). As in-camera-generated TIFF and JPEG files do notallow this approach, the following analysis assumes a completeRAW workflow, yielding completely linearly developed files inwhich the DNs are still equal to the ones initially generated bythe sensor but with all three channels completely reconstructed.

IV. ARCHAEOLOGICAL RESULTS

Do the three dissimilar spectral responses of the D50NIR

allow the researcher to gain more archaeological informationout of a straight-from-the-camera NIR frame? The answer tothis question is illustrated in Fig. 7. In the upper part [Fig. 7(a)and (b)], two 16-bit versions of the same aerial photograph areshown, taken with the D50NIR on July 20, 2007 at 13:30 habove the central Adriatic Roman town of Septempeda(43◦ 14′10′′ N, 13◦ 11′52′′ E–WGS84). Fig. 7(a) was createdby opening the original RAW file in Capture NX (NikonCorporation), a dedicated RAW converter for NEF files. Aswith all RAW converters, this program automatically appliesa tonal correction to the data (a gamma-like curve to rectifythe mismatch between the approximately logarithmic humanvisual system (HVS) and the linear sensor) and white balancesthe scene by multiplying every spectral channel with a presetweight, thereby correcting for the differential spectral response

of the DSC and compensating for the varying spectral output ofthe light source.

Fig. 7(b), on the other hand, was converted and demosaickedusing dcraw. The corresponding histogram shows that the chan-nels are not equal [unlike in Fig. 7(a)], and the maximum DNsare also smaller than the Capture NX version, indicating that thefile is completely linearly processed. Histogram stretching ofFig. 7(a), which is often necessary to tackle the nonmaximizedtonal range in NIR aerial photographs, yields the greater con-trast seen in Fig. 7(c). Although some features start to becomefaintly apparent, this result is largely inferior to Fig. 7(d), whichclearly indicates lighter and darker patches in the colza field,indicating the presence of underground structures such as roads,buildings, and ditches. The approach that yielded the result inFig. 7(d) was a simple arithmetic operation on Fig. 7(b), i.e.,

F (i, j) =[R(i, j) − G(i, j)][G(i, j) + B(i, j)]

(6)

in which F (i, j) is the final pixel, and R, G, and B indicate thevalue of this pixel in the red, green, and blue channels, respec-tively (a computation that is valid, as demosaicking attributedeach pixel with three complete spectral channels).

This operation clearly enhances the contrast between the soiland the vegetation, as well as biomass differences in the canopy,revealing subtle dissimilarities that are largely masked in thestructure of the original image [1]. The result is no coincidence.Although the bands used are rather broad (85-nm FWHM and95-nm FWHM), dividing them yields a so-called simple ratio(SR), a result that is also known as the ratio vegetation index(VI) or VI number. As the first true VI developed by Birthand McVey [85], Jordan [86], and Pearson and Miller [87], thisratio is known to indicate the amount of green biomass or leafarea index (LAI) better than either band alone [86], [88], [89].In all three of these pioneering cases, an NIR waveband wasdivided by a part of the red spectrum (740 nm/675 nm, 800 nm/675 nm, and 780 nm/680 nm, respectively). Although [16] also



Fig. 8. Comparison between (a) a conventional photograph and (b)–(d) three versions of a NIR photograph depicting approximately the same scene. (b) Thecomplete NIR frame. (c) The Blue NIR channel. (d) The result of the SR.

suggested a RNIR/R700 ratio, it was opted to divide the red bythe NIR band, just out of convenience rather than followingother scholars (e.g., [90]). This way, the resulting vegetationmarks have a greater resemblance to crop marks as they appearin the visible spectrum. Because the maxima of the red andNIR bands are situated near 730 and 815 nm, respectively, theoperation also has close resemblance to the R850/R710 ratio,with the latter being proven by Datt [76], [91] to exhibit a verystrong correlation with chlorophyll content.

In addition to these comparisons, Fig. 7(d) demonstrates thatthis simple VI is effective, exploiting the fact that when dealingwith healthy green vegetation, absorption is high in the redband, whereas the plant’s mesophyll tissue allows for a strongNIR reflection. Correspondingly, these areas are displayed darkin the output. In the case of the Roman road in the center of thepicture, the bare soil and/or decreased LAI markedly increasethe magnitude of the red/NIR ratio, creating lighter areas ornegative crop marks. Although the SPOT-3-similar blue band[92] has the advantage over the green or green + blue channelthrough not including any visible radiation, the incorporationinto the SR did not yield better results (as all pictures were takenbefore the DSC’s spectral characterization and the signal of theblue channel was not optimized to counter the noise levels).Longer exposures with a higher SNR should yield equal, if notbetter, results.

In a second example, the same SR was tested on remotelysensed data from a totally different situation. Fig. 8(a) showsthe grayscale and histogram-stretched version of a CanonEOS 300D digital color photograph of the western grasslandpart of the Italian Adriatic Roman coastal colony Potentia(43◦ 24′53′′ N, 13◦ 40′14′′ E–WGS84), taken on July 17, 2007at 15:00 h. It shows an excavation area (1), traces of the Romanstreet pattern (2), and a plot of cut grass (3), needed to performgeophysical research. Additionally, two paths to the excavationarea are depicted: one created by mowing (4) and a secondsmaller path of trampled vegetation (5) as a result of passage toand from the excavation area. Just as the traces of the wheel-barrow traffic (6), the latter is characterized by a yellowish-

brown appearance, which is a very strong visual indication ofplant stress [15]. Fig. 8(b) and (c), respectively, shows a demo-saicked, linearly converted, and histogram-stretched 16-bitaerial D50NIR photograph and its extracted blue layer, takenon the same day at 12:45 h.

Due to the extreme and long-term drought-induced stress theplants suffered from in the Italian summer of 2007, Fig. 8(b)[and certainly the pure NIR image in Fig. 8(c)] clearly showsthe traces of the Roman street pattern much better than Fig. 8(a).Although the stressed plants reflect greater green and red ra-diation (due to the substantial loss of chlorophyll), the streettraces stay faint in the visible domain as the surroundingvegetation is also wilted to a certain extent and the lowercanopy closure causes an increased reflectance due to a lowerdensity of photosynthetic pigments per unit soil surface area.Consequently, the differences between both vegetation stagesin the visible domain are small when compared to the NIRreflectance dissimilarity. The fact that these NIR crop marks areeven visible in grasslands indicates the very high soil moisturedeficits this vegetation is suffering from [6]. Moreover, colorinfrared (CIR) imaging was also reported earlier to have a clearadvantage over color photography in detecting archaeologicalcrop marks in pastures during summer [93], whereas pure NIRshould better reveal crop marks in dry vegetation [94].

On the other hand, all other features mentioned are easierto distinguish in the visible domain than in any of the D50NIR’sthree layers, as the decrease in total chlorophyll content is muchlarger than the change in the internal cellular structure of thevegetation. However, the aforementioned ratio again clearlyreveals [Fig. 8(d)] these biomass related traces—the square,both the paths, and the wheelbarrow area. As the street patternalmost completely disappears in Fig. 8(d), this feature is lessrelated to large differences in chlorophyll content and LAI.

Although both pictures were not simultaneously taken fromthe same spot (at 15:00 h from the airplane and circa 2 hbefore with the use of the Helikite—marked by its shadow inthe middle of the frame), the angles of view of the DSCs andthe position of the sun did not change to such an extent that the



Fig. 9. (a) Visible image of the central part of the Roman town of Ricina.(b) NIR image of the same scene with some contrast enhancement. (c) Outputwhen applying the SR with the channels from (b). The images were acquiredwith (a) a Nikon D200 and (b) and (c) a Nikon D50NIR.

observed differences could be attributed to them. Indeed, theparameter that changed the most was the solar geometry, whoseeffects are known to be of limited importance [95], certainlywhen the sun has a very small zenith angle [96].

In addition to negative crop marks, positive crop marks mightalso be distinguished by the SR. From the contrast-enhancedRAW image in Fig. 9(b), two zones with higher and denservegetation are obviously registered brighter when compared tothe surrounding plant canopy, due to the fact that the larger bio-mass of both features effects a higher reflection of incident NIRradiation. The visible frame from this scene [Fig. 9(a)], simul-taneously captured with the NIR image above the center of theRoman town of Ricina (43◦ 19′41′′ N, 13◦ 25′26′′ E–WGS84)on May 15, 2008 at 11:27 h, gives only a small hint of the pres-ence of these nonarchaeological positive grass marks [1 and 2in Fig. 9(a)]. Moreover, the hydrographical features visible inFig. 9(b) are largely indiscernible in Fig. 9(a), showing theimportance of NIR acquisition in this situation [19]. Notwith-

standing, the NIR record fails to clearly distinguish betweenthe stone walls of the Roman theater (upper part of the frame)and the grass growing in between. Calculating the SR yieldsFig. 9(c). When comparing all three frames, the magnitude ofreflectance dissimilarity in the grass field seems largest in theSR output. This mathematical operation also highlighted thelack of contrast between the theater walls and the vegetation,although it was not able to visualize the old hydrographicalfeatures.

V. DISCUSSION

From the results presented, it is clear that the archaeologicalpotential of a modified NIR-enabled DSC cannot be underes-timated. Both the use of individual spectral channels (e.g., thepure NIR image generated by the blue diodes) and that of arith-metic operations performed on a combination of channels (e.g.,the calculation of an SR) offer many opportunities to visuallyenhance archaeologically related anomalies and/or even revealcompletely new archaeological information (as shown in [19]and [45]). Although the application of NIR aerial imaging isby no means novel in archaeological reconnaissance, the ad-ditional advantages modified DSCs can offer in the generationand interpretation of NIR photographs are substantial. Not onlydo they significantly simplify the complete workflow, but theyalso expand the possibilities known from the film-based NIRapproach (pure NIR or CIR), without the costs of the latter.

However, the real-world examples also point to some impor-tant issues. First, both visible information and NIR information(pure NIR and calculated SR) clearly need to be used togetherto get a relevant archaeological picture [93] and in other nonar-chaeological disciplines [97], certainly at times when stresshas sufficiently developed, causing lower NIR reflectance ofthe canopy. From an interpretational point of view, the visibleinformation remains very important since the HVS is trainedto spot and interpret vegetation marks (as well as soil, shadow,and other patterns) in this part of the spectrum. Moreover, whendealing with chlorotic vegetation, reflectance data in the visibledomain are also of utmost importance as these very commonnegative crop marks are extremely hard to distinguish in a pureNIR image (as also witnessed in [98]), even though the SR cantackle this issue to a large extent. Therefore, building a simplecamera rig to hold two DSCs is advised to simultaneouslyacquire NIR and visible wavelengths (while offering the possi-bility to mathematically combine particular spectral channels).

Second, all photographs (except those in Fig. 9) were ac-quired in less-than-optimal circumstances, because long hotdry periods present the least discriminating conditions to fly in[21]. It can be expected that flying directly after rainfall couldsignificantly improve the results yielded by the D50NIR and thecalculated SR.

Third, the values of the SR sometimes exhibit very littlevariation, a phenomenon that can largely be attributed to twocauses. On one hand, the photographs under consideration showgrassland and semiarid zones, which are regions where the SRis known to be less effective in discriminating biomass/LAIvariations [99]. To counteract this, other mathematicaloperations were tried (particularly normalizations and VIs such



as difference VI (DVI) and normalized DVI). Generally, it wasthis SR that yielded the best and certainly the most consistentresults in these low-cover areas, which confirms to a degree theresults of the work of Baugh and Groeneveld [100].

On the other hand and more importantly, the applied SR doesnot really involve the mean red reflected radiant flux to meanNIR radiant flux. Whereas the blue + green channel (with aspectral range at half maximum of 780–875 nm and a sensitivitypeaking at 815 nm) is well suited as a reference band, beingvery little affected by either chlorophyll or water vapor absorp-tion [75], the red − green channel is still spectrally too broad tobe effectively used as a band that shows maximum sensitivityto pure chlorophyll absorption. Although the green subtractionproved very useful in removing much NIR radiation from thered channel, the resulting response curve—which has a spectralrange of 690–775 nm at half maximum—completely overlapsthe stress-sensitive red-edge region (i.e., the very steep increasein a healthy green plant’s reflectance curve at the edge of thevisible light and the beginning of the NIR spectrum [101]),something that should be omitted as it reduces the accuracy ofvegetation investigation [102]–[104]. A solution to tackle theseproblems of the D50NIR and the resulting SR is being workedon, involving flying with another simultaneously operated DSCthat acquires only radiation from the red-edge spectral region(690–710 nm). This zone has been proven several times to givethe most consistent leaf (and even canopy) reflectance responseto plant physiological stress [102], [105]–[110] and is thereforeof extreme importance in several narrow-band VIs for chloro-phyll estimation, even at the canopy level [111]–[113]. As thisrange is severely compromised in unmodified DSCs, a similarmodified DSC equipped with a narrow-band interference filterattached to the lens would be needed to generate aerial framesusing only the reflected radiation from this stress-sensitive sideof the chlorophyll absorption band. This would increase thecorrelation of the proposed reflectance ratio to plant senescenceand stress, allowing the spectral characteristics of the D50NIR

to be more fully exploited. Such an approach offers archaeol-ogists an affordable and easily managed multispectral tool thatcan provide useful information on the vegetation’s physiolog-ical and morphological conditions to aid in the survey of thearchaeological subsurface. If flying with a second (visible) orthird (visible and 700 nm) DSC is impossible to achieve, thespectral characteristics of the D50NIR and the resulting SR willstill most likely allow more relevant vegetation information tobe gathered in comparison with only a pure NIR band.

However, no matter how efficient and accurate this new“tool” can be, an increase in site discovery rate using multispec-tral imaging with DSCs is unlikely as long as the predominantflying strategy of “observer-directed” survey and photographyis in practice [114]. This approach generates extremely selec-tive (i.e., biased) data that are totally dependent on an airborneobserver recognizing archaeological phenomena. Thus, subsur-face soil disturbances that are visually imperceptible at the timeof flying will not make it into an NIR photograph (even if thespectral response in this domain is distinct). The large-scaleuse of the techniques advocated in this paper require a new (orcall it additional) approach to aerial archaeology, that is, flyingto collect geographically unbiased photographs of large areas

(a point that was already raised by other scholars concerningaerial imaging in the visible domain [114]–[117]). Otherwise,nonvisible and narrow-band imaging will only enhance therecord of known features and—in the best case—reveal pre-viously undetected archaeological details within a site that canbe seen from above (which, however, should still not be under-estimated, as new evidence may always alter the archaeologicalappraisal [118]).

VI. CONCLUSION

Archaeological aerial reconnaissance has long been and, to acertain extent, is still largely equated to flying around in a smallaircraft, using still cameras to record archaeological anomaliesrecognized by the airborne observer. Although satellite andmultispectral and hyperspectral airborne data have been usedin a variety of archaeological surveys, most users often lackboth the financial and staff resources to acquire and handlethe majority of these data (let alone the fact that the imageacquisition is executed without taking the specific archaeolog-ical requirements and constraints into account). This does not,however, imply that technical enhancements have to be ignoredand certainly not if they can cheaply be achieved. It is thereforeencouraging to see that the products of the current digitalphotography industry can have a great contribution in the low-cost technological improvements needed to better understandthe buried landscape record. In 1936, Reeves wrote about aerialarchaeology, pointing out that as “its methods and techniqueare improved, aerial photography will increase in scientificvalue” [119, p. 107]. Seen from this perspective, the ability ofmodified DSCs to acquire nonvisible data in wide and/or narrowwavebands can be just the tool archaeologists need to increasethe scientific value of every single flight. However, testing thesetools on their spectral capabilities is an absolute prerequisitefor the optimal use of the generated aerial (archaeological)imagery, given the fact that no two imaging matrices are alike.Once all essential characteristics are known, such highly NIR-sensitive devices provide a cheap, compact, robust, and easy-to-handle means for a “spectroscopic” aerial approach.

Allowing that the presented imagery was acquired in anunfavorable period and the red–green channel seems signifi-cantly broader than the ideal 690–710-nm band, the individualchannels of a modified Nikon D50 proved very useful in thecalculation of a simple VI to indicate chlorophyll-related issues,whereas the pure broadband NIR channels are more suitedto reveal severe drought and nutrient stress in the canopyreflectance [120]. In addition to using the three channels gen-erated by one single modified DSC, their combination withdiscrete specifically chosen spectral bands (which are generatedby a tandem of photographic cameras) looks promising. Just astheir use is not solely restricted to crop mark archaeology [19],NIR-enabled DSCs could also be applied in several nonarchae-ological domains, including agriculture, forest management,and the mapping of water bodies. Rather than making theother methods of data acquisition obsolete, modified DSCsthus offer convenient low-cost possibilities to yield essentialbeyond-visible information for the benefit of various aerial andground-based disciplines.



ACKNOWLEDGMENT

The authors would like to thank D. Cowley (Royal Commis-sion on the Ancient and Historical Monuments of Scotland) forproofreading the manuscript and the two anonymous reviewersfor their helpful comments. This paper arises from the firstauthor’s Ph.D., which was conducted with the permission ofthe Fund for Scientific Research—Flanders (FWO).

REFERENCES

[1] A. R. Beck, “Archaeological site detection: The importance of contrast,”presented at the Annu. Conf. Remote Sensing and PhotogrammetrySociety, RSPSoc, Newcastle, U.K., 2007. [Online]. Available: http://www.ceg.ncl.ac.uk/rspsoc2007/papers/189.pdf

[2] T. E. Avery and T. R. Lyons, Remote Sensing. Aerial and Terres-trial Photography for Archeologists. Washington, DC: Cultural Resour.Manage. Division, 1981.

[3] G. W. G. Allen, “Discovery from the air,” Aerial Archaeol., vol. 10,pp. 37–92, 1984.

[4] R. Evans, “The weather and other factors controlling the appearance ofcrop marks on clay and ‘difficult’ soils,” in Populating Clay Landscapes,J. Mills and R. Palmer, Eds. Stroud, U.K.: Tempus, 2007, pp. 16–27.

[5] R. Evans and R. J. A. Jones, “Crop marks and soil marks at two ar-chaeological sites in Britain,” J. Archaeol. Sci., vol. 4, no. 1, pp. 63–76,Mar. 1977.

[6] R. J. A. Jones and R. Evans, “Soil and crop marks in the recognition ofarchaeological sites by air photography,” in Aerial Reconnaissance forArchaeology, D. R. Wilson, Ed. London, U.K.: Council Brit. Archaeol.,1975, pp. 1–11.

[7] D. N. Riley, “The technique of air-archaeology,” Archaeol. J., vol. 101,pp. 1–16, 1946.

[8] D. N. Riley, “Factors in the development of crop marks,” AerialArchaeol., vol. 4, pp. 28–32, 1979.

[9] D. R. Wilson, Air Photo Interpretation for Archaeologists. Stroud,U.K.: Tempus, 2000.

[10] E. B. Knipling, “Physical and physiological basis for the reflectanceof visible and near-infrared radiation from vegetation,” Remote Sens.Environ., vol. 1, no. 3, pp. 155–159, Dec. 1970.

[11] G. S. Rabideau, C. S. French, and A. S. Holt, “The absorption andreflection spectra of leaves, chloroplast suspensions, and chloroplastfragments as measured in an Ulbricht sphere,” Am. J. Bot., vol. 33,no. 10, pp. 769–777, 1946.

[12] J. T. Woolley, “Reflectance and transmittance of light by leaves,” PlantPhysiol., vol. 47, no. 5, pp. 656–662, May 1971.

[13] M. N. Merzlyak and A. A. Gitelson, “Why and what for the leavesare yellow in autumn? On the interpretation of optical spectra of senesc-ing leaves (Acer platanoides L.),” J. Plant Physiol., vol. 145, no. 3,pp. 315–320, 1995.

[14] A. Young and G. Britton, “Carotenoids and stress,” in Stress Responsesin Plants: Adaptation and Acclimation Mechanisms, R. Hobbs andH. Mooney, Eds. New York: Springer-Verlag, 1990, pp. 87–112.

[15] M. L. Adams, W. D. Philpot, and W. A. Norvell, “Yellowness index:An application of spectral second derivatives to estimate chlorosis ofleaves on stressed vegetation,” Int. J. Remote Sens., vol. 20, no. 18,pp. 3663–3675, Dec. 1999.

[16] A. A. Gitelson, M. N. Merzlyak, and H. K. Lichtenthaler, “Detection ofred-edge position and chlorophyll content by reflectance measurementsnear 700 nm,” J. Plant Physiol., vol. 148, no. 3/4, pp. 501–508, 1996.

[17] G. A. F. Hendry, J. D. Houghton, and S. B. Brown, “Tansley reviewno. 11. The degradation of chorophyll—A biological enigma,” NewPhytol., vol. 107, no. 2, pp. 255–302, Oct. 1987.

[18] A. R. Benton, R. H. Haas, J. W. Rouse, and R. W. Toler, “Low-cost aerialphotography for vegetation analysis,” J. Appl. Photogr. Eng., vol. 2,pp. 46–49, 1976.

[19] G. J. J. Verhoeven, “Near-infrared sensing of vegetation marks,” in Be-yond Conventional Boundaries. New Technologies, Methodologies, andProcedures for the Benefit of Aerial Archaeological Data Acquisitionand Analysis, G. J. J. Verhoeven, Ed. Zelzate, Belgium: Nautilus Aca-demic Books, 2009, pp. 193–216.

[20] E. W. Chappelle, M. S. Kim, and J. E. McMurtrey, III, “Ratio analysisof reflectance spectra (RARS): An algorithm for the remote estimationof the concentrations of chlorophyll a, chlorophyll b, and carotenoidsin soybean leaves,” Remote Sens. Environ., vol. 39, no. 3, pp. 239–247,Mar. 1992.

[21] H. W. Gausman, “Leaf reflectance of near-infrared,” Photogramm. Eng.,vol. 40, pp. 183–191, 1974.

[22] H. W. Gausman, W. A. Allen, and R. Cardenas, “Reflectance ofcotton leaves and their structure,” Remote Sens. Environ., vol. 1, no. 1,pp. 19–22, Mar. 1969.

[23] V. I. Myers, M. D. Heilman, R. J. P. Lyon, L. N. Namken, D. Simonett,J. R. Thomas, C. L. Wiegand, and J. T. Woolley, “Soil, water and plantrelations,” in Remote Sensing With Special Reference to Agriculture andForestry. Washington, DC: Nat. Acad. Sci., 1970, pp. 253–297.

[24] R. C. Heller, “Imaging with photographic sensors,” in Remote SensingWith Special Reference to Agriculture and Forestry. Washington, DC:Nat. Acad. Sci., 1970, pp. 35–72.

[25] C. C. D. Lelong, P. C. Pinet, and H. Poilvé, “Hyperspectral imag-ing and stress mapping in agriculture: A case study on wheat inBeauce (France),” Remote Sens. Environ., vol. 66, no. 2, pp. 179–191,Nov. 1998.

[26] M. S. Moran, P. J. Pinter, B. E. Clothier, and S. G. Allen, “Effectof water stress on the canopy architecture and spectral indices of ir-rigated alfalfa,” Remote Sens. Environ., vol. 29, no. 3, pp. 251–261,Sep. 1989.

[27] S. H. Shakir Hanna and B. Girmay-Gwahid, “Spectral characterizationof water stress impact on some agricultural crops: II. Studies on alfalfausing handheld radiometer,” in Proc. SPIE—Remote Sensing for Agri-culture, Ecosystems, and Hydrology, Barcelona, Spain, 1998, vol. 3499,pp. 296–306.

[28] S. H. Shakir Hanna and B. Girmay-Gwahid, “Spectral characterizationof water stress impact on some agricultural crops: III. Studies on Sudangrass and other different crops using handheld radiometer,” in Proc.SPIE—Remote Sensing for Earth Science, Ocean, and Sea Ice Appli-cations, Florence, Italy, 1999, vol. 3868, pp. 154–166.

[29] O. Braasch, “Gallipoli Ahead—Air survey between the Baltic andMediterranean,” Publikácia vznikla v rámci Centra excelentnosti SAVVýskumné centrum Najstarsích dejín Podunajska pri Archeologickomústave SAV v Nitre, pp. 84–96, 2007, Nitra, Slovakia.

[30] Eastman Kodak Company, Conversion of Light (Photons) to ElectronicCharge, Aug. 5, 1999. [Online]. Available: http://www.kodak.com/US/en/digital/pdf/ccdPrimerPart1.pdf

[31] J. Nakamura, “Basics of image sensors,” in Image Sensors and SignalProcessing for Digital Still Cameras, J. Nakamura, Ed. Boca Raton,FL: Taylor & Francis, 2006, pp. 53–93.

[32] F. van de Wiele, “Photodiode quantum efficiency,” in Proc. NATOAdv. Study Inst. Solid State Imag., Louvain-la-Neuve, Belgium, 1975,pp. 47–90.

[33] F. Dierksin “Sensitivity and image quality of digital cameras,”Basler AG, Ahrensburg, Germany, Oct. 27, 2004. [Online]. Available:http://www.baslerweb.com/downloads/9835/Image_Quality_of_Digital_Cameras.pdf

[34] G. C. Holst, CCD Arrays, Cameras and Displays. Bellingham, WA:SPIE, 1996.

[35] A. J. P. Theuwissen, Solid-State Imaging With Charge-Coupled Devices.Dordrecht, The Netherlands: Kluwer, 1995.

[36] G. J. J. Verhoeven, “Imaging the invisible using modified digital stillcameras for straightforward and low-cost archaeological near-infraredphotography,” J. Archaeol. Sci., vol. 35, no. 12, pp. 3087–3100,Dec. 2008.

[37] D. D. Busch, Digital Infrared Pro Secrets. Boston, MA: Thomson,2007.

[38] T. Koyoma, “Optics in digital still cameras,” in Image Sensors and SignalProcessing for Digital Still Cameras, J. Nakamura, Ed. Boca Raton,FL: Taylor & Francis, 2006, pp. 21–51.

[39] S. F. Ray, Applied Photographic Optics. Lenses and Optical Systemsfor Photography, Film, Video, Electronic and Digital Imaging. Oxford,U.K.: Focal Press, 2002.

[40] Eastman Kodak Company, Kodak KAF-8300CE Image Sensor.3326 (H) x 2504 (V). Full-Frame CCD Color Image Sensor withSquare Pixels for Color Cameras, Apr. 18, 2005. [Online]. Available:http://www.kodak.com/ezpres/business/ccd/global/plugins/acrobat/en/datasheet/fullframe/KAF-8300LongSpec.pdf

[41] D. M. Chabries, S. W. Booras, and G. H. Bearman, “Imaging thepast: Recent applications of multispectral imaging technology to de-ciphering manuscripts,” Antiquity, vol. 77, no. 296, pp. 359–372,Jun. 2003.

[42] D. Har, Y. Son, and S. Lee, “SLR digital camera for forensic pho-tography,” in Proc. SPIE—Sensors and Camera Systems for Scientific,Industrial, and Digital Photography Applications V, San Jose, CA, 2004,vol. 5301, pp. 276–284.

[43] D. King, P. Walsh, and F. Ciuffreda, “Airborne digital frame cameraimaging for elevation determination,” Photogramm. Eng. Remote Sens.,vol. 60, no. 11, pp. 1321–1326, 1994.



[44] D. J. King, K. Haddow, and D. G. Pitt, “Evaluation of CIR digital cameraimagery in regeneration assessment: Research design and initial results,”in Proc. 1st North Amer. Symp. Small Format Aerial Photography,Cloquet, MN, 1997, pp. 186–195.

[45] G. J. J. Verhoeven, “Becoming a NIR-sensitive aerial archaeologist,” inProc. SPIE—Remote Sensing for Agriculture, Ecosystems, and Hydrol-ogy IX, Florence, Italy, 2007, vol. 6742, pp. 674 20Y-1–674 20Y-13.

[46] J. Chen, Digital Infrared at Jim Chen Photography, 2007. [Online].Available: http://www.jimchenphoto.com

[47] The Nikon Guide to Digital Photography With the D50 Digital Camera,Nikon Corporation, Tokyo, Japan, 2005.

[48] “ICX413AQ. APS Size Diagonal 28.4 mm (Type 1.8) 6.15 M Effec-tive Pixels Color CCD Image Sensor.,” Cx-News. Sony SemiconductorNews, 2002. [Online]. Available: http://www.sony.net/Products/SC-HP/cx_news/vol28/pdf/icx413np.pdf

[49] B. E. Bayer, “Color imaging array,” U.S. Patent 3 971 065, Jul. 20, 1976.[50] K. Parulski and K. Spaulding, “Color image processing for digi-

tal cameras,” in Digital Color Imaging Handbook, G. Sharma, Ed.Boca Raton, FL: CRC Press, 2003, pp. 728–757.

[51] P. G. Dorrell, Photography in Archaeology and Conservation.Cambridge, U.K.: Cambridge Univ. Press, 1994.

[52] B. Rørslett, All You Ever Wanted to Know About Digital UV and IRPhotography, But Could Not Afford to Ask, 2004. [Online]. Available:http://www.naturfotograf.com/UV_IR_rev00.html

[53] G. J. J. Verhoeven and J. Loenders, “Looking through black-tintedglasses—A remotely controlled infrared eye in the sky,” in 2nd Int. Conf.Remote Sens. Archaeol. Proc. 2nd Int. Workshop—From Space to Place,Rome, Italy, 2006, pp. 73–79.

[54] G. J. J. Verhoeven, J. Loenders, F. Vermeulen, and R. Docter, “Helikiteaerial photography or HAP—A versatile means of unmanned, radiocontrolled low altitude aerial archaeology,” Arch. Prosp., vol. 16, no. 2,pp. 125–138, 1994.

[55] B. Rørslett, Lens Survey and Subjective Evaluations, 2007. [Online].Available: http://www.naturfotograf.com/lens_surv.html

[56] T. Mizoguchi, “Evaluation of image sensors,” in Image Sensorsand Signal Processing for Digital Still Cameras, J. Nakamura, Ed.Boca Raton, FL: Taylor & Francis, 2006, pp. 179–203.

[57] R. Berry and J. Burnell, The Handbook of Astronomical Image Process-ing. Richmond, VA: Willmann-Bell, 2005.

[58] G. J. J. Verhoeven, “It’s all about the format—Unleashing the power ofRAW aerial photography,” Int. J. Remote Sens., to be published.

[59] C. J. Skinner and L. Bergeronin “Characteristics of NICMOS detectordark observations,” Space Telesc. Sci. Inst., Baltimore, MD, Oct. 1997.[Online]. Available: http://www.stsci.edu/hst/nicmos/documents/isrs/isr_026.pdf

[60] H. T. Hytti, “Characterization of digital image noise properties based onRAW data,” in Proc. SPIE—Image Quality and System Performance III,2005, vol. 6059, p. 605 90A.

[61] Eastman Kodak Company, CCD Image Sensor Noise Sources,Jan. 10, 2005. [Online]. Available: http://www.kodak.com/global/plugins/acrobat/en/digital/ccd/applicationNotes/noiseSources.pdf

[62] Y. Reibel, M. Jung, M. Bouhifd, B. Cunin, and C. Draman, “CCDor CMOS camera noise characterisation,” Eur. Phys. J., Appl. Phys.,vol. 21, no. 1, pp. 75–80, Jan. 2003.

[63] C. Buil, Comparaison du Canon 10D et du Nikon D70 en imagerie as-tronomique longue pose. [Online]. Available: http://astrosurf.com/build/70v10d/eval.htm

[64] S. B. Howell, Handbook of CCD Astronomy. Cambridge, U.K.:Cambridge Univ. Press, 2006.

[65] R. Wodaski, The New CCD Astronomy: How to Capture the Stars Witha CCD Camera in Your Own Backyard. Duvall, WA: New AstronomyPress, 2002.

[66] J. Lukás, J. Fridrich, and M. Goljan, “Digital camera identification fromsensor pattern noise,” IEEE Trans. Inf. Forensics Secur., vol. 1, no. 2,pp. 205–214, Jun. 2006.

[67] W. Budde, “Definition of the linearity range of Si photodiodes,” Appl.Opt., vol. 22, no. 11, pp. 1780–1784, Jun. 1983.

[68] P. L. Vora, J. E. Farrell, J. D. Tietz, and D. H. Brainard, “Linear modelsfor digital cameras,” in Proc. IS&T’s 50th Annu. Conf.—A Celebrationof All of Imaging, Cambridge, MA, 1997, pp. 377–382.

[69] R. N. Clark, The Nikon D50 Digital Camera: Sensor Noise, DynamicRange, and Full Well Analysis, Mar. 3, 2006. [Online]. Available: http://clarkvision.com/imagedetail/evaluation-nikon-d50/index.html

[70] Leofoo, Additional Information on Nikkor 20 mm Ultra-WideangleLenses—20 mm f/3.5, 2001. [Online]. Available: http://www.mir.com.my / rb / photography / companies / nikon / nikkoresources / late70nikkor /ultrawides/20mma.htm

[71] A. Mann, Infrared Optics and Zoom Lenses. Bellingham, WA: SPIE,2000.

[72] Nikon Corporation, AF Nikkor 50 mm f/1.8D, 2007. [Online]. Avail-able: http://www.nikonimaging.com/global/products/lens/af/normal/af_50mmf_18d/index.htm

[73] J. Peñuelas, I. Filella, C. Biel, L. Serrano, and R. Savé, “The reflectanceat the 950–970 nm region as an indicator of plant water status,” Int. J.Remote Sens., vol. 14, no. 10, pp. 1887–1905, Jul. 1993.

[74] P. S. Thenkabail, R. B. Smith, and E. De Pauw, “Hyperspectral vege-tation indices and their relationships with agricultural crop characteris-tics,” Remote Sens. Environ., vol. 71, no. 2, pp. 158–182, Feb. 2000.

[75] C. J. Tucker, “A comparison of satellite sensor bands for vege-tation monitoring,” Photogramm. Eng. Remote Sens., vol. 44, no. 11,pp. 1369–1380, 1978.

[76] B. Datt, “Visible/near infrared reflectance and chlorophyll content ineucalyptus leaves,” Int. J. Remote Sens., vol. 20, no. 14, pp. 2741–2759,Sep. 1999.

[77] J. Adams, K. Parulski, and K. E. Spaulding, “Color processing in digitalcameras,” IEEE Micro, vol. 18, no. 6, pp. 20–30, Nov./Dec. 1998.

[78] D. H. Brainard and D. Sherman, “Reconstructing images fromtrichomatic samples: From basic research to practical applications,” inProc. 3rd IS&T/SID Color Imag. Conf.—Color Sciences, Systems andApplications, Scottsdale, AZ, 1995, pp. 4–10.

[79] L. Chang and Y.-P. Tan, “Effective use of spatial and spectral correlationsfor color filter array demosaicking,” IEEE Trans. Consum. Electron.,vol. 50, no. 1, pp. 355–365, Feb. 2004.

[80] B. C. de Lavarène, D. Alleysson, and J. Hérault, “Practical implementa-tion of LMMSE demosaicing using luminance and chrominance spaces,”Comput. Vis. Image Underst., vol. 107, no. 1/2, pp. 3–13, Jul. 2007.

[81] B. K. Gunturk, J. Glotzbach, Y. Altunbasak, R. W. Schafer, andR. M. Mersereau, “Demosaicking: Color filter array interpolation. Ex-ploring the imaging process and the correlations among three colorplanes in single-chip digital cameras,” IEEE Signal Process. Mag.,vol. 22, no. 1, pp. 44–54, Jan. 2005.

[82] R. Lukac and K. N. Plataniotis, “Demosaicked image postprocessingusing local color ratios,” IEEE Trans. Circuits Syst. Video Technol.,vol. 14, no. 6, pp. 914–920, Jun. 2004.

[83] K. Hirakawa and T. W. Parks, “Adaptive homogeneity-directeddemosaicing algorithm,” IEEE Trans. Image Process., vol. 14, no. 3,pp. 360–369, Mar. 2005.

[84] D. Coffin, Decoding RAW Digital Photos in Linux, 2008. [Online]. Avail-able: http://cybercom.net/~dcoffin/dcraw/

[85] G. S. Birth and G. R. McVey, “Measuring the color of growing turfwith a reflectance spectrophotometer,” Agron. J., vol. 60, pp. 640–643,1968.

[86] C. F. Jordan, “Derivation of leaf-area index from quality of light on theforest floor,” Ecology, vol. 50, no. 4, pp. 663–666, 1969.

[87] R. L. Pearson and L. D. Miller, “Remote mapping of standing cropbiomass for estimation of the productivity of the short-grass prairie,Pawnee National Grasslands, Colorado,” in Proc. 8th Int. Symp. RemoteSens. Environ., 1972, pp. 1355–1379.

[88] J. Peñuelas and I. Filella, “Visible and near-infrared reflectance tech-niques for diagnosing plant physiological status,” Trends Plant Sci.,vol. 3, no. 4, pp. 151–156, 1998.

[89] M. Schlerf, C. Atzberger, and J. Hill, “Remote sensing of forest biophys-ical variables using HyMap imaging spectrometer data,” Remote Sens.Environ., vol. 95, no. 2, pp. 177–194, Mar. 2005.

[90] D. Zhao, K. R. Reddy, V. G. Kakani, J. J. Read, and S. Koti, “Selectionof optimum reflectance ratios for estimating leaf nitrogen and chloro-phyll concentrations of field-grown cotton,” Agron. J., vol. 97, no. 1,pp. 89–98, 2005.

[91] B. Datt, “A new reflectance index for remote sensing of chlorophyllcontent in higher pants: Tests using eucalyptus leaves,” J. Plant Physiol.,vol. 154, no. 1, pp. 30–36, 1999.

[92] G. Begni, “Selection of the optimum spectral bands for the SPOT satel-lite,” Photogramm. Eng. Remote Sens., vol. 48, no. 10, pp. 1613–1620,1982.

[93] J. N. Hampton, “An experiment in multispectral air photography forarchaeological research,” Photogramm. Rec., vol. 8, no. 43, pp. 37–64,Apr. 1974.

[94] R. Lasaponara and N. Masini, “Detection of archaeological crop marksby using satellite QuickBird multispectral imagery,” J. Archaeol. Sci.,vol. 34, no. 2, pp. 214–221, Feb. 2007.

[95] D. B. Lobell, G. P. Asner, B. E. Law, and R. N. Treuhaft, “Viewangle effects on canopy reflectance and spectral mixture analysis ofconiferous forests using AVIRIS,” Int. J. Remote Sens., vol. 23, no. 11,pp. 2247–2262, Jun. 2002.



[96] D. F. Wanjura and J. L. Hatfield, “Vegetative and optical character-istics of four-row crop canopies,” Int. J. Remote Sens., vol. 9, no. 2,pp. 249–258, Feb. 1988.

[97] R. N. Colwell, “Spectrometric considerations involved in making ruralland use studies with aerial photography,” Photogrammetria, vol. 20,no. 1, pp. 15–33, Feb. 1965.

[98] J. H. Everitt, D. E. Escobar, J. R. Noriega, M. R. Davis, andI. Cavazos, “A digital imaging system and its application to naturalresource management,” in Proc. 1st North Amer. Symp. Small FormatAerial Photography, Cloquet, MN, 1997, pp. 123–135.

[99] A. R. Huete, K. Didan, T. Miura, E. P. Rodriguez, X. Gao, andG. Ferreira, “Overview of the radiometric and biophysical performanceof the MODIS vegetation indices,” Remote Sens. Environ., vol. 83,no. 1/2, pp. 195–213, Nov. 2002.

[100] W. M. Baugh and D. P. Groeneveld, “Broadband vegetation index per-formance evaluated for a low-cover environment,” Int. J. Remote Sens.,vol. 27, no. 21, pp. 4715–4730, Nov. 2006.

[101] D. N. H. Horler, M. Dockray, and J. Barber, “The red edge of plantleaf reflectance,” Int. J. Remote Sens., vol. 4, no. 2, pp. 273–288,1983.

[102] G. A. Carter, “Ratios of leaf reflectances in narrow wavebands as indica-tors of plant stress,” Int. J. Remote Sens., vol. 15, no. 3, pp. 697–703,1994.

[103] L. S. Galvao, I. Vitorello, and M. A. Pizarro, “An adequate band posi-tioning to enhance NDVI contrasts among green vegetation, senescentbiomass, and tropical soils,” Int. J. Remote Sens., vol. 21, no. 9,pp. 1953–1960, Jun. 2000.

[104] C. J. Tucker, “Red and photographic infrared linear combinations formonitoring vegetation,” Remote Sens. Environ., vol. 8, no. 2, pp. 127–150, May 1979.

[105] G. A. Carter, “Responses of leaf spectral reflectance to plant stress,”Am. J. Bot., vol. 80, no. 3, pp. 239–243, Mar. 1993.

[106] G. A. Carter, “General spectral characteristics of leaf reflectanceresponses to plant stress and implications for the remote sensing of veg-etation,” in Proc. ASPRS Annu. Conf.—Gateway to the New Millennium,St. Louis, MO, 2001. CD-ROM.

[107] G. A. Carter, W. G. Cibula, and R. L. Miller, “Narrow-band re-flectance imagery compared with thermal imagery for early detec-tion of plant stress,” J. Plant Physiol., vol. 148, no. 5, pp. 515–522,1996.

[108] G. A. Carter and L. Estep, “General spectral characteristics of leaf re-flectance responses to plant stress and their manifestation at the land-scape scale,” in From Laboratory Spectroscopy to Remotely SensedSpectra of Terrestrial Ecosystems, R. S. Muttiah, Ed. Dordrecht,The Netherlands: Kluwer, 2002, pp. 271–293.

[109] G. A. Carter and R. L. Miller, “Early detection of plant stress by dig-ital imaging within narrow stress-sensitive wavebands,” Remote Sens.Environ., vol. 50, no. 3, pp. 295–302, Dec. 1994.

[110] A. A. Gitelson and M. N. Merzlyak, “Signature analysis of leafreflectance spectra: Algorithm development for remote sensing ofchlorophyll,” J. Plant Physiol., vol. 148, no. 3/4, pp. 494–500, 1996.

[111] N. C. Coops, C. Stone, D. S. Culvenor, L. A. Chisholm, andR. N. Merton, “Chlorophyll content in eucalypt vegetation at the leafand canopy scales as derived from high spectral resolution data,” TreePhysiol.: Int. Botanical J., vol. 23, no. 1, pp. 23–31, Jan. 2003.

[112] C. P. Ferri, A. R. Formaggio, and M. A. Schiavinato, “Narrow bandspectral indexes for chlorophyll determination in soybean canopies[Glycine max (L.) Merril],” Braz. J. Plant Physiol., vol. 16, no. 3,pp. 131–136, Dec. 2004.

[113] H. K. Lichtenthaler, A. A. Gitelson, and M. Lang, “Non-destructivedetermination of chlorophyll content of leaves of a green and anaurea mutant of tobacco by reflectance measurements,” J. Plant Physiol.,vol. 148, no. 3/4, pp. 483–493, 1996.

[114] R. Palmer, “If they used their own photographs they wouldnot take them like that,” in From the Air: Understanding Aerial Archae-ology, K. Brophy and D. Cowley, Eds. Stroud, U.K.: Tempus, 2005,pp. 94–116.

[115] S. Coleman, “Taking advantage: Vertical aerial photographs commis-sioned for local authorities,” in Populating Clay Landscapes, J. Millsand R. Palmer, Eds. Stroud, U.K.: Tempus, 2007, pp. 28–33.

[116] J. Mills, “Bias and the world of the vertical aerial photograph,” inFrom the Air: Understanding Aerial Archaeology, K. Brophy andD. Cowley, Eds. Stroud, U.K.: Tempus, 2005, pp. 117–126.

[117] R. Palmer, “Seventy-five years v. ninety minutes: Implications ofthe 1996 Bedfordshire vertical aerial survey on our perceptions ofclayland archaeology,” in Populating Clay Landscapes, J. Mills andR. Palmer, Eds. Stroud, U.K.: Tempus, 2007, pp. 88–103.

[118] J. N. Hampton, “Aerial reconnaissance for archaeology: Uses of thephotographic evidence,” Photogramm. Rec., vol. 9, no. 50, pp. 265–272,Oct. 1977.

[119] D. M. Reeves, “Aerial photography and archaeology,” Amer. Antiq.,vol. 2, no. 2, pp. 102–107, Oct. 1936.

[120] J. Dungan, L. Johnson, C. Billow, P. Matson, J. Mazzurco, J. Moen, andV. Vanderbilt, “High spectral resolution reflectance of Douglas fir grownunder different fertilization treatments: Experiment design and treatmenteffects,” Remote Sens. Environ., vol. 55, no. 3, pp. 217–228, Mar. 1996.

Geert J. Verhoeven was born in 1978. He receivedthe Master’s degree in archaeology from Ghent Uni-versity, Ghent, Belgium, in 2002. Since 2003, he hasbeen working at the Department of Archaeology andAncient History of Europe, Ghent University. FromSeptember 2004 till October 2008, he was a Ph.D.fellowship of the Research Foundation—Flanders(FWO) and developed new technologies, methodolo-gies, and data processing procedures for the benefitof aerial archaeological data acquisition and analysis.For this research, he obtained the Ph.D. degree in

May 2009.Since 2003, he has been working at the Department of Archaeology and

Ancient History of Europe, Ghent University. His main research interestsconcern remote sensing technology, GIS, aerial and ground-based photography,photogrammetry, and archaeological computing.

Philippe F. Smet was born in 1979. He receivedthe M.Sc. and Ph.D. degrees in physics from GhentUniversity, Ghent, Belgium, in 2001 and 2005,respectively.

He is currently a Postdoctoral Researcher for theFund for Scientific Research—Flanders (FWO) withLumiLab, Department of Solid State Sciences, GhentUniversity. His main research is focused on colorconversion materials for light-emitting diodes andpersistent luminescent materials for safety applica-tions. His other research topics include the effects of

particle size on the emission properties of rare-earth-doped materials.

Dirk Poelman was born in 1963. He received thePh.D. degree in physics, on electroluminescent thinfilms, from Ghent University, Ghent, Belgium.

He is currently leading the research group Lu-miLab, Department of Solid State Sciences, GhentUniversity. In addition, he lectures several courseson bachelor and master levels. He is a coauthorof more than 130 international publications andconference contributions. His research interests in-clude luminescent powders and thin films, structuralcharacterization of materials using microscopic and

X-ray techniques, and photocatalysis for air purification.

Frank Vermeulen was born in 1960. He received thePh.D. degree in archaeology from Ghent University,Ghent, Belgium, in 1988.

Since 1999, he has been a Full-Time Professorin Roman archaeology and archaeological methodswith the Department of Archaeology and AncientHistory of Europe, Ghent University. His researchmainly focuses on the archaeology of landscapes,with an emphasis on Mediterranean environmentsand the development of geoarchaeological method-ology and fieldwork. He has organized seven inter-

national congresses, published more than ten archaeological monographs, andwritten more than 80 articles in international journals and series.


spectral characterization of a digital still camera’s nir … › ~pfsmet › papers › a1_33...

Documents