You are on page 1of 16

See discussions, stats, and author profiles for this publication at: http://www.researchgate.


A Quadratic Model with Nonpolynomial Terms
for Remote Colorimetric Calibration of 3D Laser
Scanner Data Based on Piecewise Cubic
Hermite Polynomials
Impact Factor: 0.76 · DOI: 10.1155/2015/606948



Massimiliano Guarneri

Massimo Francucci







M. Ferri de Collibus

Giorgio Fornetti







Available from: Alessandro Danielis
Retrieved on: 10 October 2015

and there are an increasing number of applications that would benefit from the addition of calibrated intensity data to the topographic information. in producing faithful 3D colour models of real targets and classifying easier and more reliable automatic tools. The development of the most widespread instrumental survey techniques (topography and stereophotogrammetry) has been considerable and has made both the acquiring and the processing of a large amount of data possible.2 Massimiliano Guarneri. 14 pages http://dx. and angle of incidence to the reflective surface.2 Mario Ferri De Received 2 June 2015. Fermi and Massimiliano Guarneri. the purely geometric information provided by the vast majority of currently available scanners is not enough for most applications.2 and Arianna Mencattini1 1 Department of Electronic Engineering.2 Giorgio Fornetti. This is an open access article distributed under the Creative Commons Attribution License. varies with distance. for example. 1. While studies of TLS radiometric calibration are sparsely available. for example.danielis@libero. in making object recognition. 00133 Rome. such as accuracy and detection range. Using piecewise cubic Hermite polynomials. The intensity information from terrestrial laser scanners (TLS) has become an important object of study in recent years [7. distribution. Three-dimensional (3D) laser scanning represents one of the ultimate steps of the technological progress in the field of morphological survey and it offers new and important opportunities [6].org/10. The quality of a rendered laser image strongly depends on the accuracy of the calibration procedure used. Revised 21 July 2015. Frascati.Hindawi Publishing Corporation Mathematical Problems in Engineering Volume 2015. massimiliano. Energy and Sustainable Economic Development. which permits unrestricted use.2 Massimo Francucci. Results concerning laser scanner digitizations of artworks are reported to confirm the effectiveness of the method. National Agency for New Technologies. Accurate calibrated intensity could give added value for laser scanning campaigns. 8]. Article ID 606948. In cultural heritage area. Italy ENEA.1155/2015/606948 Research Article A Quadratic Model with Nonpolynomial Terms for Remote Colorimetric Calibration of 3D Laser Scanner Data Based on Piecewise Cubic Hermite Polynomials Alessandro Danielis.doi. Via E. Introduction An active area of research in computer graphics involves the processing of colorimetric data with established and innovative mathematical methods [1–5]. and reproduction in any medium. research on remote colorimetric calibration started in the last . The processing of intensity data from terrestrial laser scanners has attracted considerable attention in recent years.guarneri@enea. Generally. to the best of the authors’ knowledge.1. Accurate calibrated intensity could give added value for laser scanning campaigns. Via del Politecnico 1. Such distinguishing colour information opens new scenarios and problems for remote colorimetry. “Tor Vergata” University of Rome. ale. Colorimetric data recorded by the prototype on certified diffusive targets is processed for generating a remote Lambertian model used for assessing the accuracy of the proposed algorithm. The scanner response may be subject to an input-output nonlinearity that needs to be modeled. a quadratic model with nonpolynomial terms for reducing inaccuracies occurring in remote colour measurement is implemented. where indeed accurate colorimetric data is needed. in generating faithful 3D colour models of real targets. laser scanner performance. provided the original work is properly cited. and in classifying easier and more reliable automatic tools. This paper presents a remote calibration method for self-registered RGB colour data provided by a 3D tristimulus laser scanner prototype. object reflectivity. Accepted 27 July 2015 Academic Editor: Erik Cuevas Copyright © 2015 Alessandro Danielis et al. 00044 Rome. Italy 2 Correspondence should be addressed to Alessandro Danielis.

g. Ideally. and remote diagnosis purposes. calibrated RGB data is merged with range information to produce high-quality 3D digital models. because of the optoelectronic system features.1. This capability makes colour information as important as range data. This implies that an accurate calibration with a certified reference target is necessary for producing RGB-ITR colorimetric models. These calibrations have assumed a linear relation between the red. without being affected by scanning geometry measurement. Such distinguishing feature opens new scenarios and a new class of problems for remote colorimetry [14]. green. The improvement gives added value for laser scanning campaigns. Differently from commercial 3D scanners. The present study shows that. independently on the conditions under which those regions were sounded during the scan. Linear calibration of gray targets is performed by calculating the ratio between the three RGB signals acquired by the instrument for each target and the amplitude value on the three calibration curves selected by the corresponding distance 𝑑. dissemination. The effects of distance and target reflectance on the recorded intensity along with the development of the proposed correction method are analyzed. The RGB-ITR colour measuring capabilities are influenced by various factors. a patented 3D colour laser scanner prototype completely designed and realized at ENEA laboratories of Frascati (Rome) for cultural heritage purposes. a correct colour calibration procedure should represent the content of a real target 3D model by reflecting the intrinsic properties of the target surface. The on-focus distance for these curves corresponds to 𝑑0 = 4. 514 nm. their characteristic bell-shape. and big surfaces at short and large distances from the scanner. and 440 nm) and is able to simultaneously collect colour and structure information for any investigated sampled surface point in a working range of about 3–30 meters. which consists in illuminating a movable white target with the laser beam at fixed steps (e. Figures 1 and 2 report the calibration curves and the detected gray target signals as functions of distance at different ranges from the scanner. due to nonlinearities of the optoelectronic system. so the collected power also falls down. well described in [13]. respecting CIE standards. Note that. or blue light power reflected by a particular surface. are the result of the optical parameters tuned up for maximizing the signal-noise ratio of the detected voltages and their shape is independent from distance. Linear Calibration. The model is capable of representing real. providing meaningful and more accurate 3D RGBITR colour images. For 𝑑 < 𝑑0 . In this study. 10 cm) along the scanning range. The implementation is performed with a fast algorithm that uses a quadratic model with nonpolynomial terms and piecewise cubic Hermite polynomials with first-order accurate derivative approximation method. A remote Lambertian model for assessing the accuracy of the algorithm and several experimental results that demonstrate the effectiveness provided by the novel calibration procedure are reported. two colorimetrically indistinguishable regions of the surface should be represented in the model as having the same colour.. that is. As a consequence. A standardization of colour information detected by RGB-ITR system. since the generation and superimposition of texture images are realised via software and commercial cameras. Namely. providing intensity data much more suitable for cataloguing. restoration. where approximately all the RGB curves have a maximum. The purely geometric output provided by the majority of currently available laser scanners is not enough for most cultural heritage applications [13].2 few years with the RGB-ITR (Red Green Blue-Imaging Topological Radar) system [9]. For each channel 𝜆 𝑖 the measured reflectance factor of a gray target is given by gray 𝑅𝑑 gray (𝜆 𝑖 ) = 𝑉𝑑 𝑉𝑑white (𝜆 𝑖 ) (𝜆 𝑖 ) ⋅ 𝑅white (𝜆 𝑖 ) . it is not possible to obtain the same RGB signals on certified Lambertian targets at all the distances of measurement. The advantages of 3D imaging systems in the field of cultural heritage are now recognized and widely accepted [10–12]. 2. Mathematical Problems in Engineering Raw colour data returned by the RGB-ITR simply consists of a triplet of voltage values. The curves exhibit the same behaviour. 16] of three monochromatic sources (660 nm. For 𝑑 > 𝑑0 the collected power falls down mainly because of the misalignment of the optical axes and the 1/𝑑2 typical trend [13]. 2. as collected by the receiving optics and revealed by the detector. and blue back-reflected signals and scene reflectances at each distance of measurement. These raw detected data form the calibration curves. that is. ambient light effects. all the calibration curves. green. This technique has many drawbacks and limitations in terms of achievable visual quality and accuracy. was already performed in the past with colour calibrations by distance information and by MINOLTA spectrophotometer-CM-2600d [9]. the ENEA prototype can simultaneously record for each investigated point range and RGB colour data. Each value represents the red. and so forth. RGB-ITR Colour Detection The RGB-ITR system is a tristimulus laser scanner prototype based on the amplitude modulation [15. small. the effects produced by the nonlinearities of the optoelectronic system are illustrated by calibrating certified Lambertian gray targets of 50% and 20% reflectances with raw colour measurements performed on a white diffusive target at different distances from the system. this assumption produces undesired effects in the remote colour measurement of certified diffusive gray targets (Labsphere Inc. this trend is counterbalanced by the fact that the receiver intercepts only part of the reflected light.) and proposes a nonlinear calibration for compensating the input-output characteristics of the device. Finally. they all have a maximum that can be fixed by varying some optical parameters. A common practice is to superimpose textures derived from ordinary digital photos onto the 3D models generated by most devices. (1) .4 m (curves in Figure 1) and 𝑑0 = 20. The RGB triplets are then processed with a calibration procedure.8 m (curves in Figure 2).

01 0. and 99% nominal reflectance have been used as reference targets for the calculations.02 0.04 0.02 0.045 0.015 0.01 0. Four Spectralon (Labsphere Inc.015 0.015 0. linear calibration (1) does not return the same reflectance values for light and dark gray targets at each distance.05 0. 50%. the white target used in this measurement.01 0.03 0. Methods This section presents the mathematical models introduced in the proposed calibration procedure.005 0.005 0 2 3 4 5 Distances (m) 6 7 0 2 3 4 5 Distances (m) 6 7 White raw data 50% gray raw data 20% gray raw data White raw data 50% gray raw data 20% gray raw data (a) Red channel amplitudes (b) Green channel amplitudes 0.04 0. the punctual colour model of the same object at different distances is not exactly represented as having the same colour.8 meters.) Diffusive Reflectance Standards (SRS) with 2%.005 0 2 3 4 5 Distances (m) 6 7 White raw data 50% gray raw data 20% gray raw data (c) Blue channel amplitudes Figure 1: RGB signals measured on 99%. Figure 3 shows the result of the linear calibration procedure applied to light gray and dark gray targets at 39 sampled distances. ranging from 2. As shown by calibrated signals in Figure 3. and 20% reflectance targets at 39 sampled distances from the scanner.025 0. 50%.02 0. gray where 𝑅𝑑 (𝜆 𝑖 ) is the measured reflectance factor at the distance 𝑑 and 𝑅white (𝜆 𝑖 ) is the certified spectral reflectance factor of the perfect diffuser.025 0. 20%.3 0. .03 0.03 0. that is.8 to 6. that is.035 0. This nonlinearity generates discontinuities in the colour representation of gray targets. 3.045 0.04 0.035 Voltages (mV) Voltages (mV) Mathematical Problems in Engineering 0.035 Voltages (mV) 0.025 0.

and education purposes.4 Mathematical Problems in Engineering 0.01 0 17 0. Quadratic Calibration Model.05 Voltages (mV) Voltages (mV) 0.1. 3. More accurate data is much more suitable for restoration and diagnosis aids.02 0.02 0.07 0. while adaptation to human perception makes RGB-ITR images more interesting for dissemination. cataloguing.04 18 19 20 21 Distances (m) 22 23 White raw data 50% gray raw data 20% gray raw data 0 18 17 19 20 21 Distances (m) 22 23 White raw data 50% gray raw data 20% gray raw data (a) Red channel amplitudes (b) Green channel amplitudes 0.07 Voltages (mV) 0. 19]: 𝑌 1/3 { 116 ⋅ ( ) − 16 { { 𝑌𝑛 𝐿∗ = { { 𝑌 {𝑙 ⋅ ( ) 𝑌𝑛 { 𝑌 6 3 >( ) 𝑌𝑛 29 𝑌 6 3 if <( ) 𝑌𝑛 29 if (2) .05 0.05 0. and 20% reflectance targets at 24 sampled distances from the scanner.07 0.1.06 0.06 0.08 0.08 0. To address the second constraint. ranging from 17.03 0. As shown by linear calibration results in Section 2.04 0. the back-reflected RGB signals of light gray and dark gray targets are not constant with distance. the nonlinear term is defined to be comparable to the power function defined by the Munsell equation of lightness 𝐿∗ according to the standard CIE 1931 and 1964 [18.01 0. A model that considers different certified Lambertian reflectances is needed for compensating this system nonlinearity and for producing a more accurate colorimetric representation of real scenes.06 0. 50%.7 to 22.3 meters.04 0.03 0.01 0 17 18 19 20 21 Distances (m) 22 23 White raw data 50% gray raw data 20% gray raw data (c) Blue channel amplitudes Figure 2: RGB signals measured on 99%. A nonlinear term is introduced in the model in order to both eliminate inaccuracies in the remote colour measurement of gray targets and mimic the nonlinear perceptual response to luminance of human vision [17].02 0.03 0.

The model coefficients are computed by constructing and solving a set of simultaneous equations. and 660 nm.2 0. Given a set of 𝑁 calibration distances 𝑑1 .55 Light gray target Dark gray target Reflectance 0. 𝐾= ∑𝜆 𝑆 (𝜆 𝑖 ) ⋅ 𝑦cmf (𝜆 𝑖 ) ⋅ Δ𝜆 (4) For each distance of measurement and for each channel a curve is constructed with a quadratic model with nonpolynomial terms necessary for interpolating at each distance four points given by lightness values measured with the MINOLTA spectrophotometer-CM-2600d on four SRS. Namely. and 𝑧cmf (𝜆)). 𝑑𝑁.Mathematical Problems in Engineering 5 0. 𝑎2 .45 0. Δ𝜆 is the measurement wavelength interval.3 0. 𝑇2. ∑𝜆 represents summation across wavelength 𝜆. respectively. 𝑅(𝜆) is the object’s spectral reflectance factor. and 𝐾 is a conventional normalizing constant defined as [19] 100 . . The term 𝑌/𝑌𝑛 is called the luminance factor. the tristimulus value 𝑌 for an object is the luminance factor and can be approximated as 𝑌 = 𝐾 ⋅ ∑𝑆 (𝜆 𝑖 ) ⋅ 𝑅 (𝜆 𝑖 ) ⋅ 𝑦cmf (𝜆 𝑖 ) Δ𝜆. By definition. when illuminated and viewed under specified geometric conditions. the model 𝑦(𝑡) should be in the form 𝑦 (𝑡) = 𝑎0 + 𝑎1 ⋅ 𝑓 (𝑡) + 𝑎2 ⋅ 𝑓 (𝑡) ⋅ 𝑡 + 𝑎3 ⋅ 𝑓 (𝑡) ⋅ 𝑡2 . (5) where 𝑓(𝑡) is the nonpolynomial function of the model and 𝑎0 . given generic 𝑡 data. 𝑎1 . and 𝑓(𝑉) is the nonlinear term defined as 𝑓 (𝑉) = 𝑉1/3 (7) comparable to the lightness measure in (2). the ratio between the luminances of a specimen and of a perfect diffuser.5 0. with 𝑙 = (29/3)3 . . that is. 514 nm. The MINOLTA measurements were performed in a spectrum interval of 380 nm ÷ 740 nm with Δ𝜆 = 10 nm assuming the 𝐷65 illuminant and the 10∘ CIE standard observer.4 0. . related to 440 nm. can be well approximated with the spectrophotometer measurements. The 2% diffusive target is assumed to be the zero (black) for the calibration measurements. 𝑦cmf (𝜆) is one of the three CIE standard observer colour-matching functions (𝑥cmf (𝜆). 𝑎3 are the model coefficients. RGB-ITR lightnesses. RGB calibrated values for light gray and dark gray targets at 39 sampled distances. the model used for each channel and for a generic distance 𝑑𝑘 can be defined with the following compact form: 3 𝑖−1 𝑦𝑇1 = 𝑎0 + 𝑓 (𝑉𝑇1 ) ⋅ ∑𝑎𝑖 ⋅ (𝑉𝑇1 ) 𝑖=1 3 𝑖−1 𝑦𝑇2 = 𝑎0 + 𝑓 (𝑉𝑇2 ) ⋅ ∑𝑎𝑖 ⋅ (𝑉𝑇2 ) 𝑖=1 3 (6) 𝑖−1 𝑦𝑇3 = 𝑎0 + 𝑓 (𝑉𝑇3 ) ⋅ ∑𝑎𝑖 ⋅ (𝑉𝑇3 ) 𝑖=1 3 𝑖−1 𝑦𝑇4 = 𝑎0 + 𝑓 (𝑉𝑇4 ) ⋅ ∑𝑎𝑖 ⋅ (𝑉𝑇4 ).35 0.65 0. 𝑎0 . 𝑎1 . 𝑦cmf (𝜆). 𝑎2 .6 Three-channel reflectances White target 0. and 𝑎3 are the unknown model coefficients. The use of four targets and the pursuit of the dual objective mentioned above require the model to be nonlinear in the coefficients and in the data. Since reflectance values of gray targets are almost constant in the visible spectrum. 𝑉 is the input measured voltage. 𝑦 is the imposed measured lightness. 𝜆 (3) where 𝑆(𝜆) is a CIE illuminant.25 0. This is accomplished by forming . 𝑇3. and 𝑇4 correspond to the four Lambertian targets used in the calibration procedure. . 𝑖=1 where 𝑇1.15 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 Distance samples 2 3 4 5 Distance (m) 6 7 Red channel Green channel Blue channel (a) Linear calibrated colour images (b) RGB reflectance plots as function of distance Figure 3: Linear calibration results.

.8 0.01 0. but at the same time they are too powerful. . Interpolation with Piecewise Cubic Hermite Polynomials. ℎ𝑁−1 with (10) As an example.4 ) . a 𝐶1 cubic Hermite spline is defined by a set of polynomials ℎ0 . They are simple to calculate. 𝑥𝑁) with 𝑎 = 𝑥0 < 𝑥1 < ⋅ ⋅ ⋅ < 𝑥𝑁 = 𝑏. Scene intensity matrix V.2 . the approximation can only be affected in the two intervals [𝑥𝑖−1 . a function 𝑓 : [𝑎. .6 Mathematical Problems in Engineering The implementation of the proposed model is given in Algorithm 1. Input. scene distance matrix D. and a set of partition points 𝑥⃗ = (𝑥0 . 𝐻(𝑥) also has a local extremum. For generic 𝑇 target and 𝑑 distance. Algorithm 1 (main algorithm steps of the model). .025 Linear model Imposed data Nonlinear model Figure 4: Linear model. and imposed targets lightnesses yi at calibration distance 𝑑𝑖 are as follows: (V) 󳨀→ (V)𝑛×𝑚×3 (D) 󳨀→ (D)𝑛×𝑚 (d) 󳨀→ (d)1×𝑘 (xi )1×4 = (𝑥𝑖. Given an interval [𝑎. The evaluation of the design matrix at scene detected voltages Vs for each RGB channel gives the nonlinear calibrated intensities Ic . calibration distance vector d. (8) The least-square solution of the following system returns the model coefficients at the distance 𝑑: (𝑎) [𝑋] = (𝑦) .015 0. These properties make monotonic Hermite interpolation extremely local: if a single interpolation point 𝑥𝑖 changes.02 0. (13) . (9) This procedure is repeated for all 𝑁 distances. 𝑥𝑖 ] and [𝑥𝑖 . . 𝑥𝑖. at a generic distance of measurement and with 25 mV of detected signal measured on the white target. 𝑏] → R. Differently. the nonlinear model for the red channel. . ℎ𝑖 (𝑥𝑖 ) = 𝑓 (𝑥𝑖 ) ℎ𝑖 (𝑥𝑖+1 ) = 𝑓 (𝑥𝑖+1 ) Input voltages (mV) 𝑉𝑇1/3 3. with derivative 𝑓󸀠 : [𝑎. 𝑥𝑖+1 ] which share that partition point. nonlinear model. Robust and efficient interpolation of the model coefficients at scene distances is needed for an accurate calibration measure. . a design matrix [𝑋]. on intervals where the data are monotonic. 𝑏]. .2 . 𝐻(𝑥) is also monotonic. where each column represents a variable used to predict the response (a term in the model) and each row corresponds to one observation of those variables.6 0. ℎ𝑖󸀠 (𝑥𝑖 ) = 𝑓󸀠 (𝑥𝑖 ) (11) ℎ𝑖󸀠 (𝑥𝑖+1 ) = 𝑓󸀠 (𝑥𝑖+1 ) for 𝑖 = 0. 𝑖=0 (12) The Hermite form has two control points and two control tangents for each polynomial. is shown in Figure 4. 𝑏] → R. 𝑥1 . in terms of time required to determine the interpolant and to evaluate it. . This means that. the RGB signals can be very high (tens of mV) and the black signal can be used in the calibration measure since it can be greater than or equal to about 1 mV. ℎ1 . the approximated lightnesses. Calibrated intensities 1 0. and the imposed lightness at a generic distance. that is.2 0 0 0. Consider the following.005 0. measured targets intensities xi . Using piecewise cubic Hermite polynomials. 𝑦𝑖. near the maximum of the calibration curves. An example of Hermite interpolation of generic data is shown in Figure 5.1 . 𝑁 − 1. Input voltages are expressed in mVolts. 𝑥𝑖. of the scene: [Ic ] = (a) [XVs ] . At this distance the black target detected signal is less than 10 𝜇V (comparable to the instrument noise signal). . 21]. the matrix is given by [𝑋𝑇 ] = [1 𝑉𝑇 ⋅ 𝑉𝑇1/3 𝑉𝑇2 ⋅ 𝑉𝑇1/3 ] .2.3 .1 . 𝑥𝑖. 𝑦𝑖. 4 × 𝑁 coefficients are interpolated at the acquired scene distances. 𝑦𝑖. The slopes at 𝑥𝑗 are chosen in such a way that 𝐻(𝑥) preserves the shape of the data and respects monotonicity [20]. thus it does not affect the calculation.3 .4 ) (yi )1×4 = (𝑦𝑖. . Piecewise cubic Hermite interpolating polynomials 𝐻(𝑥) have properties that meet these requests [20. at points where the data has a local extremum. The spline formed by this collection of polynomials can be defined as 𝑛 𝐻 (𝑥) = ∑𝑓 (𝑥𝑖 ) ℎ𝑖 (𝑥) + 𝑓𝑖󸀠 ℎ𝑖󸀠 (𝑥) . 1. .4 0.

3 𝑖. 𝑔𝑖 = V𝑖+1 − V𝑖−1 . Calibrated intensity matrix Ic is as follows: (Ic ) 󳨀→ (Ic )𝑛×𝑚×3 . thus. Hermite interpolations are performed with both approximation methods on data unevenly perturbated by adding normally distributed random numbers on acquired distances.2 ] [ =[ ]. The more accurate approximation method performs Hermite interpolation by using (21) and . (14) (1) Form the design matrix at distance 𝑑𝑖 : [Xi ]4×4 1/3 1/3 1/3 2 𝑥𝑖.1 𝑎𝑖. .2 𝑎𝑖.2 𝑖. .3 𝑖.5 ÷ 7. that is. the derivative values of each polynomial are approximated in such a way that the interpolation could not be significantly affected by errors that can occur in the calibration procedure.4 𝑥𝑖. Numerical simulations are performed on single channel data detected in a range of [2. V𝑛 ).1 ⋅ 𝑥𝑖.2 ⋅ 𝑥1/3 𝑥2 ⋅ 𝑥1/3 ] 𝑖.4 ] = (2) Compute model coefficients (Ai )1×4 (𝑎𝑖.1 1 𝑥𝑖.1 [ ] [1 𝑥1/3 𝑥𝑖. they are not useful in this context. (19) (Ic )(𝑛⋅𝑚)×3 󳨀→ (Ic )𝑛×𝑚×3 .4 ⋅ 𝑥𝑖.2 𝑖. .2 𝑖. . the derivatives 𝑔 on the left and right edges are computed by taking the forward differences: 𝑔1 = V2 − V 1 𝑑2 − 𝑑1 V − V𝑛−1 𝑔𝑛 = 𝑛 . In (b) the finer Hermite interpolation is highlighted. (20) (3) Repeat all for each of the k sampling distances: (4) Interpolate model coefficients at scene distances corresponding to the elements of (D)𝑛×𝑚 with piecewise cubic Hermite polynomials within (d)1×𝑘 and (A)𝑘×4 : (5) Evaluate the model for each element of the vectorized (V)(𝑛⋅𝑚)×𝑗 intensity channel: (6) Reshape the calibrated scene intensity matrix: In this work.3 ] (15) 1/3 1/3 1/3 2 [1 𝑥𝑖.4 ) with least-square method at distance 𝑑𝑖 : (Ai )1×4 [Xi ]4×4 = (yi )4×1 . . Giving a set of 𝑛 calibration distances (𝑑) = (𝑑1 . In order to compare the robustness of this gradient computation method to errors occurring in remote measurements with respect to a more accurate one. . (18) (Ic )1×1 = (XV )1×4 (A)4×1 .7 f(x) f(x) Mathematical Problems in Engineering x True Data x True Data Spline Hermite interpolation (a) Spline Hermite interpolation (b) Figure 5: An example of data interpolation with cubic Hermite polynomials and spline. More accurate methods for finite difference approximation of derivatives that distinguish between evenly and unevenly spaced points do not work well with noisy data.1 ⋅ 𝑥𝑖.5] m for 500 different sets of random perturbations ranging from 0 to 5 cm.4 𝑥𝑖. (17) (Ai )𝑘×4 󳨀→ (Ai )(𝑛⋅𝑚)×4 .4 ⋅ 𝑥𝑖.3 [ 𝑖.3 𝑎𝑖. Output. . [1 𝑥1/3 𝑥 ⋅ 𝑥1/3 𝑥2 ⋅ 𝑥1/3 ] 𝑖. 𝑑𝑛 ) and detected voltages (V) = (V1 . wrong distance measurements. 𝑑𝑛 − 𝑑𝑛−1 (21) On interior points V𝑖 with 1 < 𝑖 < 𝑛 taking the central differences. 𝑑𝑖+1 − 𝑑𝑖−1 (22) This simple approximation is first-order accurate for evenly and unevenly spaced points.3 𝑖. Three different sets of perturbations are shown in Figure 6. (16) (Ai )1×4 󳨀→ (Ai )𝑘×4 . noise is introduced in the calibration distance measurements.1 𝑥𝑖. .

A histogram of each set is plotted along the diagonal of the figure. In the 𝑖th row. (b) perturbations p1. and p3 plotted as function of calibration distances. .8 Mathematical Problems in Engineering 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 0 1 2 3 4 0 5 1 2 3 4 5 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 0 1 2 3 4 0 5 1 2 3 4 5 5 5 5 4 4 4 3 3 3 2 2 2 1 1 1 0 0 0 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 p2 (cm) p1 (cm) (a) 5 4 3 2 1 0 2 3 4 5 6 Distances (m) 7 8 2 3 4 7 8 7 8 5 4 3 2 1 0 5 6 p3 (cm) Distances (m) 5 4 3 2 1 0 2 3 4 5 6 Distances (m) (b) Figure 6: (a) Three sets of normally distributed random perturbations. Values range from 0 to 5 cm. p2. 𝑗th column of the figure is a scatter plot of the 𝑖th set against the 𝑗th set of perturbations.

in some cases. . In (c) the mean interpolation error produced by 500 different random perturbation sets on true data is reported for both methods.08 0.06 0. the maximum perturbation introduced is of 5 cm on distance measurements.01 0 2 3 True data Perturbated data 4 5 Distances (m) 6 7 8 0 2 Method 1 Method 2 3 4 5 Distances (m) 6 7 8 Method 1 Method 2 True data Perturbated data (a) (b) ×10 9 −3 Mean interpolation error 8 7 6 5 4 3 2 1 0 0 100 200 300 Simulation number 400 500 Method 1 Method 2 (c) Figure 7: Hermite interpolation on two differently perturbated single channel pieces of data with first-order accurate (method 1) and second-order accurate (method 2) derivative approximations. the less accurate approximation method for derivatives has a low mean interpolation error that is almost constant in all the simulations.04 0.03 0.05 0. Hermite polynomials with firstorder accurate method keep good capabilities for interpolation and also exhibit low computational complexity.03 0. the more accurate one produces large interpolation errors as can be observed by Figures 7(b) and 7(c). Remote Lambertian Model for Experimental Validation.01 0. 3.9 0. this approximation is second-order accurate for both evenly and unevenly spaced coordinates.04 0. Some results of Hermite interpolation with both methods are shown in Figure 7.05 Voltages (mV) Voltages (mV) Mathematical Problems in Engineering 0. An accurate quantitative analysis for RGB-ITR colorimetric calibrations can be carried out by comparing prototype remote measurements with spectrophotometer measurements on certified Lambertian targets.02 0. Thus. Contrarily. In both cases.07 0.02 0. Contrarily to (22). As can be noted.3.06 0.07 0.08 0. (22) for evenly spaced points and replaces (22) for unevenly spaced points with the following formula: 𝑔𝑖 = V𝑖+1 ⋅ ℎ𝑖 /ℎ𝑖+1 − V𝑖−1 ⋅ ℎ𝑖+1 /ℎ𝑖 1 1 + V𝑖 ⋅ ( − ) ℎ𝑖 + ℎ𝑖+1 ℎ𝑖 ℎ𝑖+1 (23) with ℎ𝑖 = 𝑑𝑖+1 − 𝑑𝑖 .

015 0. A comparison of these results with MINOLTA measurements performed on light and dark gray targets is reported in Table 1. Linear and nonlinear calibrations of light gray SRS raw model give the images shown in Figure 9. A quantitative result of this correction is given with the RGB histograms shown in Figure 10. results concerning the application of the model on two different digitizations are reported: (i) The fresco of “Amore e Psiche” (Villa Farnesina. interpolating the detected RGB signals of the gray targets at acquired RGB-ITR real scene distances (Figure 8) and reshaping the computed vector quantities to scene matrix dimensions. caused by nonlinearities of the system. where 𝑛 is the number of sampled distances. since the small variations of the collected back-reflected signals are more visible in a neutral (whitish or grayish) region than in a colourful environment and show that two objects having similar colour properties but located at different distances from the scanner.748 ± 0.49 0. colour texture images for both targets are generated. RGB linear calibrated values are more different between them with respect to the reflectances measured by MINOLTA. The wrong gray shades on the white wall of the linear calibrated image are strongly reduced with the proposed correction method.7] m. are represented with different colours in noncalibrated data. Each RGB signal is collected by moving the targets at sampling distances of about 10 cm in a range of [2.5 ÷ 7.749 ± 0. More precisely. the white statue appears greenish.752 ± 0. Using Hermite polynomials and first-order accurate derivative approximations.560 ± 0. In this paper. Nonlinear calibration ensures that RGB values measured on gray targets are much more similar to each other and to MINOLTA measurements. remote colour models for light gray (50% reflectance) and dark gray (20% reflectance) SRS are estimated.47 𝜆 = 660 nm 0.9 to 7. three RGB [𝑛 × 1] vector quantities.5 250 1000 2000 3000 4000 4 Figure 8: Laboratory distance image.10 Mathematical Problems in Engineering 7.02 0. the nonlinear one does not show discontinuities. that is. The histograms are computed using the probability normalization function with bin width of 0. can be reduced.01 0.75 discontinuities. The model is based on the RGB measures of the four SRS. placed at different distances and illuminated by RGB-ITR lasers.505 ± 0.5 200 5 4. Applications of the Model The nonlinear model was tested with success on various RGBITR scans. The scene distances range from 3. Figure 11 reports raw data normalized by the maximum RGB values and linear and nonlinear calibrated images for the real scene acquired in laboratory at distances shown in the range image (Figure 8) used for remote Lambertian models generation. that is. On the contrary.5 50 7 6. This quantitative result demonstrates that the proposed model provides significantly better colour mapping in comparison with the conventional linear method when illuminating certified targets. (ii) The Vault of the Sistine Chapel (Vatican Museums. The raw input image is not colorimetrically correct since each RGB sampled surface point is dependent on the artwork/scanner distance and on the level of the RGB signals that illuminate each point. Instrument MINOLTA [𝑅] RGB-ITR [𝑅] mean value MINOLTA [𝐿∗ ] RGB-ITR [𝐿∗ ] mean value 𝜆 = 440 nm 0. Differently from the linear calibrated image. Rome) located at 17-18 meters of distance from the scanning system. These interpolated matrix quantities represent the remote Lambertian models. To confirm the effectiveness of the proposed colorimetric calibration. the same object is correctly represented as having the same colour at all scene distances. in Figure 9(a). which are then calibrated.7 meters.5 100 6 150 5.05 for both the Lambertian models.012 0. such as wall and statue.75 0. Rome) located at a distance of 7–9 meters from the scanning system. a comparison with the corresponding normalized raw data and linear calibrated models is shown in Figures 12 and 13.02 0. In particular. it is possible to identify objects located at different distances.011 0.75 0. In this way. The presented laboratory test was processed with controlled conditions act to simulate all the possible power variations of the laser sources equipped inside the RGB-ITR scanner. Three RGB curves as a function of distance are computed for each target. A remote Lambertian model is computed in laboratory for assessing the accuracy of the proposed algorithm (Algorithm 1). The proposed scene with large whitish areas well fit the purpose of this validation. while the small object in the lower left side of the scene is represented to be too dark and is difficult to recognize. For illustrative purposes. 4.521 ± 0. colour Table 1: Numerical comparison of the reflectance [𝑅] and lightness [𝐿∗ ] values measured by MINOLTA spectrophotometer and RGBITR with linear and nonlinear calibration methods for the 50% reflectance gray target. only some portions of these huge digitizations are reported.46 𝜆 = 514 nm 0. It can be observed that the raw images are darker than the calibrated ones and .

6 Calibrated intensity 0. Each coloured histogram corresponds to one type of RGB channel data.2 Calibrated intensity (a) 0.6 0. as quantitatively demonstrated on certified targets.3 Linear calibration 0.3 0. the linear one. related to higher red and blue ITR signals that illuminate the scene at those distances. as performed by linear calibrations. the position of the RGB scene signals in the calibration curves. but. 0. The processed colour images.6 0.2 0.1 0 0. the nonlinear calibrated models can be considered suitable colour images for cultural heritage purposes.6 0. that is. The nonlinear model is introduced in the calibration procedure .1 0 0.7 0.4 Nonlinear calibration Linear calibration 0. and approximated lightness space.7 0.4 0.5 0. It is important to underline that linear and nonlinear models correspond to two different colour spaces: approximated reflectance space.8 1 (b) Figure 10: Histograms of the calibrated dark gray (a) and light gray (b) Lambertian colour models. The problem of linear calibration for gray diffusive certified targets is discussed.8 1 0 0 0. Conclusion This paper has presented a nonlinear model for remote calibration of laser colour images. 5. demonstrate that the nonlinear model can be viewed as a colorimetric transformation that visibly reduces inaccuracies occurring when linearly calibrating grayish/whitish areas and eliminates the dependence on RGB signals’ level. the nonlinear one represents a much more accurate laser colour model. in union with the experimental validation in Section 3.3.2 Nonlinear calibration 0. Overall.4 0. Both images appear to be visually pleasing to the observer. As confirmed by experts in cultural heritage sector.2 0.5 Normalized counts Normalized counts Figure 9: Linear and nonlinear calibrations of light gray remote Lambertian model.Mathematical Problems in Engineering 11 50 50 100 100 150 150 200 200 250 250 500 500 1000 1500 2000 2500 3000 3500 4000 4500 (a) Linear calibrated remote Lambertian model 1000 1500 2000 2500 3000 3500 4000 4500 (b) Nonlinear calibrated remote Lambertian model 0. these results show that the proposed nonlinear calibration works well for real scenes located at different distances from the scanner. have higher red and blue intensity components. The model is calculated at the RGB-ITR scene distances shown in Figure 11(c).4 0. the nonlinear one.

. (a) Raw colour data (b) Linear calibrated (c) Nonlinear calibrated (d) Raw colour data (e) Linear calibrated (f) Nonlinear calibrated Figure 12: RGB-ITR raw data and linear and nonlinear calibrated colour textures of real scene at 7–9 meters of distance.12 Mathematical Problems in Engineering 50 50 100 100 150 150 200 200 250 250 1000 2000 3000 1000 4000 (a) Raw colour data 2000 3000 4000 (b) Linear calibration 50 100 150 200 250 1000 2000 3000 4000 (c) Nonlinear calibration Figure 11: Laboratory real scene raw data and calibrations.

Papanikolaou. 1999. pp. Article ID 480523. no. 10. that is. and H.” Mathematical Problems in Engineering. “Radiometric calibration of terrestrial laser scanners with external reference targets. Results that highlight the effectiveness of the proposed model in producing meaningful images for dissemination. This represents an improvement especially in the cultural heritage domain. B.-H. References [1] A. vol. Plataniotis. pp. 393–402. Joel Trussell. Acha. “Adaptive filters for color image processing. where the colorimetric characterisation of an artwork has the same significance of the structural determination of the artwork itself. 4 pages. An experimental validation of the proposed algorithm is performed with a remote Lambertian model computed by using monotonic piecewise cubic Hermite polynomials and certified diffusive targets. vol. [7] S.Mathematical Problems in Engineering 13 (a) Raw colour data (b) Linear calibrated (c) Nonlinear calibrated (d) Raw colour data (e) Linear calibrated (f) Nonlinear calibrated Figure 13: RGB-ITR raw data and linear and nonlinear calibrated colour textures of real scene at about 17-18 meters of distance. 2009. Venetsanopoulos. “Colorimetric restoration of digital images. 6.” International Journal of Advanced Research in Computer Science and Software Engineering. “Mathematical methods applied to digital image processing. vol. Professor Paolucci. G. Conflict of Interests The authors declare that there is no conflict of interests regarding the publication of this paper.” IEEE Transactions on Image Processing. C. M. “Geometric calibration of a terrestrial laser scanner with local additional parameters: .-H. Rout. 2014. N. Haring. L. Dorninger. and A. 2014. Altunbasak and H. no. [8] D. 7. Baciu. Chao. and N. July 2007. education. 4. A. 2001. In conclusion. and P. [3] V. 328–337. the proposed nonlinear correction suggests that an optimal calibration is possible for the RGB-ITR. Nicolini. and H. This quantitatively demonstrates that the nonlinear model is much more accurate than the linear one when performing remote colorimetric measurements on different certified Lambertian targets.-P. 1. pp. and cataloguing purposes are reported. as a postprocessing step for the correction of raw colorimetric data. Pfeifer. Although remote colour calibration procedure is still object of active research. Garc´ıa-San-Miguel and J. and C. Color Image Processing with Biomedical Applications. Hermite polynomials with first-order accurate derivative approximation formula have shown robust and efficient interpolation performances in presence of added noise. Khandual. The nonlinearity of the model is the consequence of the processing made on data collected by illuminating four different calibrated targets. pp. this work has shown that the processing of colorimetric data permits generating more accurate texture images that if combined with 3D laser models can provide high-quality information. Fan. Professor Santamaria. for both reducing inaccuracies occurring in remote measurement of gray targets and providing meaningful colour images for cultural heritage purposes. Serrano. P. Kaasalainen. [6] N. 144–158. Krooks. C.” Remote Sensing. no. pp. and Professor Di Pinto for their cooperation as cultural heritage experts and for having allowed working in the Sistine Chapel. K. Acknowledgments The authors thank the Vatican in the person of Mons. 529–538. “Colorimetric processing of digital colour image. 393–402. 3. [5] H. N. Lerma. 2011. 3. Chen. [2] Y. vol. A. no. “Investigating terrestrial laser scanning intensity data: quality and functional relations. Kukko. [4] R. A. Society of Photo-Optical Instrumentation Engineers (SPIE).” in Proceedings of the International Conference on Optical 3-D Measurement Techniques VIII. Kaartinen. 3. Liu. vol. 2007. Rangayyan.” Mathematical Problems in Engineering.

Higham. 2005. vol. vol. S. 17. 22.” IEEE Signal Processing Magazine.” Journal of Archaeological Science.” Advances in Optical Technologies. B. and A. and A. “Amplitude-modulated laser imager. pp.. Billmeyer and Saltzman’s Principles of Color Technology. Guarneri. Soardo. vol. A. Wiley. R. A. 1–129. Mencattini. M. J.” SIAM Journal on Numerical Analysis. with applications to ODE plotting. M. Katsev. 2014. fotometria. Fornetti. 43. Gerbino et al. M. M. 381–388. S. colorimetria e norme internazionali. Fiorentini. and A. R. vol.” Journal of Computational and Applied Mathematics. Brain.” in Proceedings of the Firenze 1999 and Venezia 2000. Francucci. S. 1999-2000. Luo. 2012. 3874–3892. 46. vol. no. “The structure and properties of color spaces and the representation of color images. Mullen. Article ID 512902. Ed. Ricci. D. L. M. 1. Springer. Cyprus. Prikhach. Ricci.CONA VIII. A. Lemessos. 2010. 2004. pp. Zha. 1992. 6 pages. 122–136. R. vol. De Collibus. C. “Remote colorimetric and structural diagnosis by RGB-ITR color laser scanner prototype. vol. Carlson. November 813. pp. pp.” ISPRS Journal of Photogrammetry and Remote Sensing. no. “RGB-ITR: an amplitude-modulated 3D colour laser scanner for cultural heritage applications . A. F. E. Santamaria. M. Bacci. Fornetti. 19. “3D remote colorimetry and watershed segmentation techniques for fresco and artwork decay monitoring and preservation. De Collibus et al. Fritsch and R. 2009. Ciatti. M. D. M. SIAI Center. 2010. Germany. Concannon. “Monotonic piecewise cubic interpolation. “The measurement and use of registered reectance and range data in scene analysis. Pelagotti. 6436 of Lecture Notes in Computer Science.” in Proceedings of the International Conference LA. 2. 2012.” in Digital Heritage: Third International Conference. W. Zege. Danielis.” Applied Optics. vol. Li. 4. M. Dubois. no. pp. Molinas. “Image processing for the analysis and conservation of paintings: opportunities and challenges. Oleari. Barni.. 2008. E. Piva. T. Video. no. Battaglino. Guarneri. E. A. 3. EuroMed 2010. “Colorimetria e Beni Culturali. pp. 287–294. Berlin. 2000. 2013. 79. pp. Misurare il colore. 238–246. Francucci.14 [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] Mathematical Problems in Engineering an automatic strategy. Duda. 1976. De Dominicis. and Multimedia Processing. 182–190.. Fisiologia della visione a colori. F. Laux. and R. G. L. vol. Nitzan. F. “Monotone piecewise cubic interpolation. September 2009. pp. no. Berns. B. and H. 39. F.” SRI Project. and P. De Collibus. U. L. “3D digitization and its applications in cultural heritage. M. P. N. Lasers in the Conservation of Artworks. M. I. 1980. and R.” Synthesis Lectures on Image. Hoepli. Stanford Research Institute. Nuvoli. 141–144. Proceedings. 5. G. .

com Abstract and Applied Analysis Volume 2014 Hindawi Publishing Corporation Hindawi Publishing Corporation Volume 201 Discrete Dynamics in Nature and Society Volume 201 Hindawi Publishing Corporation Volume 201 International Journal of Mathematics and Mathematical Sciences Mathematical Problems in Engineering Journal of Mathematics Hindawi Publishing Corporation Volume 2014 Advances in Decision Sciences Hindawi Publishing Corporation http://www.Advances in Operations Research Hindawi Publishing Corporation Volume 201 Hindawi Publishing Corporation Hindawi Publishing Corporation Volume 2014 Journal of Probability and Statistics Volume 2014 The Scientific World Journal Hindawi Publishing Corporation Volume 2014 Journal of Applied Mathematics Algebra Hindawi Publishing Corporation Mathematical Physics Hindawi Publishing Corporation International Journal of Advances in Combinatorics Hindawi Publishing Corporation Volume 2014 International Journal of Differential Equations Hindawi Publishing Corporation Journal of Function Spaces Hindawi Publishing Corporation Volume 2014 Volume 201 Submit your manuscripts at Hindawi Publishing Corporation Volume 2014 International Journal of Journal of Stochastic Analysis Optimization Hindawi Publishing Corporation Volume 201 Volume 2014 Volume 201 Discrete Mathematics Journal of Volume 2014 Hindawi Publishing Corporation Volume 2014 Journal of Complex Analysis Hindawi Publishing Corporation Volume 2014 Hindawi Publishing Corporation http://www.hindawi.