RGB-D-λ: 3D Multispectral Acquisition with Stereo RGB Cameras

Alain Trémeau

, Simone Bianco

and Raimondo Schettini

Laboratoire Hubert Curien UMR 5516, University Jean Monnet, 18 Rue Benoit Lauras, Saint-Etienne, France

DISCo – Dept. of Informatics, Systems and Communication, University of Milano-Bicocca, Milano, Italy

Keywords: Recovery Reflectance Spectra, Color Calibration, 3D Reconstruction.

Abstract: In this paper we report some key questions related to the acquisition of accurate color images adressed by

the Colour and Space in Cultural Heritage (COSCH) project. We summarize the main criteria defined by

COSCH Working Group 1 and used by the authors to help Digital Cultural Heritage (DCH) users to have a

deepen knowledge and understanding of the constraints, preconditions and practical aspects related to the

use of a multi-spectral acquisition system. We also report how color data can be estimated from spectral

data and discuss several issues related to the measurement of reflective surfaces by 2D imaging systems.

1 INTRODUCTION

The paper introduces some key questions concerning

the acquisition of accurate color images in Digital

Cultural Heritage (DCH) identified through the

“Colour and Space in Cultural Heritage” project (see

www.COSCH.info), a trans-domain European Action

(TD1201) supported, since 2013, by the European

Cooperation in Science and Technology. The results

reported in this paper were done in collaboration with

the COSCH Working Group 1 (WG1) and will be

included in the interim report of this working group

which will be published on the next COSCH bulletin

(see http://cosch.info/publications/). This work is part

of a larger project in which the colorimetric accuracy

of traditional devices characterized with state-of-the-

art techniques (Bianco, Gasparini et al., 2008,

Bianco, Schettini et al., 2009, Bianco, Bruna et al.

2011) is compared with multispectral devices

(Bianco, Colombo et al., 2011).

The activities of WG1 are mainly focused on the

“identification, characterization and testing of

spectral imaging techniques in the visible and near

infrared (IR) field” and on the “identification,

characterization and testing of imaging techniques

beyond the visible and short wave radiation” for the

study and documentation of CH artefacts. Until now,

most of the activities supervised by the WG1 have

been dedicated to the use of spectral imaging

techniques to the investigation of artworks, mainly 2-

D objects such as paintings. Spectral imaging

systems are in general grouped in multi-band, multi-

spectral and hyper-spectral systems depending on the

number of bands selected over a given spectral

interval and on their bandwidths. In the period 1989-

2004, several projects (such as CRISATEL (Ribes,

2005), VASARI (Saunders, 1993) or NARCISSE

(Lahanier, 2003)), or reports (e.g. see (Liang, 2004,

Imai, 2010)) focused on the use of 2-D multi-band

and 2-D multi-spectral imaging systems, shown the

importance of spectral information to characterize 2-

D cultural heritage objects in the visible field. In the

last decade new developments in optical techniques

led to novel ways of splitting the visible radiation

(light). The COSCH WG1 decided in 2013 to explore

the limits and advantages of the actual instruments in

these wavebands. This WG1 defined a “Round Robin

Test” (RRT) in order to compare color and

spectroscopic measurements, as well as information

on calibrated standards and laboratory mock-ups

obtained through the use of diverse imaging devices

or prototypes, specifically designed for high-quality

digital imaging of artworks and for conservation

purposes, developed by the different research groups

that participate in the COSCH Action. In their

preliminary works, WG1 members agreed that there

are still no well-established and commonly accepted

criteria and guidelines for the selection and the use of

one of the above mentioned imaging systems. In this

paper, we summarize the main criteria defined by

COSCH WG1 and used by the authors to help

Cultural Heritage users to have a deepen knowledge

and understanding of the constraints, preconditions

Trémeau, A., Bianco, S. and Schettini, R.

RGB-D-λ: 3D Multispectral Acquisition with Stereo RGB Cameras.

DOI: 10.5220/0005845406890696

In Proceedings of the 11th Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2016) - Volume 4: VISAPP, pages 689-696

ISBN: 978-989-758-175-5

689

and practical aspects related to the selection and the

use of a multi-spectral acquisition system. Next, we

report how color data can be estimated from spectral

data and discuss several issues related to the

measurement of reflective surfaces by 2D imaging

systems.

2 ACCURACY OF 3D COLOR

ACQUISITIONS

In (Boochs, 2015) the authors reported how to

perform color acquisition using a multispectral

system and what are the effects due to parametric

factors (e.g. exposure time, camera response, etc.)

and other external factors (e.g. spectral distribution of

the lighting source) influencing the accuracy of color

acquisition. They also reported how to evaluate the

accuracy of a multispectral system and how to

improve this accuracy by color calibration using a

standardized color chart (e.g. the color checker

chart). Other papers also addressed these issues, such

as (Lopez-Alvarez, 2009, De Lasarte Rigueiro,

2009). Nevertheless there is still no well-established

and commonly accepted criteria and guidelines for

the selection and the use of standardized color charts

for accurate color calibration of multispectral

imaging systems. The number of color samples, the

color gamut, the spectral distribution and the

reflectance property of these samples are the four

main factors influencing the accuracy of the

calibration process. It is also important to remind that

the color rendering of digitized images depends also

of many other parameters, such as viewing geometry,

lighting geometry, camera settings, etc. (Faridul,

2015). As example, in Figure 1, due to shadows,

highlights, surfaces reflectance, viewing geometry,

camera settings, color filters, etc. the scene digitized

has different color appearances. As a consequence,

for each viewing geometry used (i.e. each 2D color

acquisitions) we suggest to apply again color

calibration. As example, the scene shown in Figure 1

was acquired from 8 viewing positions (see Figure

2). In order to digitize this 3D scene with accurate

color values, one solution consists: first to select

sample colors on different object surfaces (as shown

in Figure 3); next to calibrate each 2D view from the

expected color values (using a sparse approach); and

lastly to refine the color calibration (i.e. to minimize

color differences) by color correction (using a dense

approach). We suggest to use the solution proposed

by H.S. Faridul et al. to minimize color errors

between 2D views with high overlap (De Lasarte

Rigueiro, 2009).

The results of the calibration process depends

also of the material property of the color chart used,

thus it is possible to obtain different values with one

color checker chart or with another one. We

experimented this with the color checked chart of the

Round Robin Test, and two other color checker

charts (used at Saint-Etienne and Milano).

To evaluate color differences between measured

values and expected values, users can use the

CIELAB ΔE metric or similar metrics such as the

CIEDE2000. To evaluate spectral differences

between measured (or estimated) spectral values and

expected values, users can use the mean value and

the standard deviation of difference values between

spectral data. Another option consist to compute the

percentage of the Integrated Irradiance Error metric

(IIE(%)) which is a relative measurement of the

(a) Same scene viewed under four viewing positions and acquired with a RGB CANON camera sensor with a red filter (wide band filter)

put in front of the camera.

(b) Same scene viewed under four viewing positions and acquired with another RGB CANON camera sensor.

Figure 1: Example of a complex scene (composed of 3D objets put on and in front of colored planar surfaces) digitized by

two cameras under different viewing positions.

RGB-SpectralImaging 2016 - Special Session on RBG and Spectral Imaging for Civil/Survey Engineering, Cultural, Environmental,

Industrial Applications

690

Figure 2: Screenshot of the dense reconstruction (922.000 points – 185.000 faces) of the 3D scene by the software Agisoft.

The 8 positions of cameras used are illustrated by blue and pink squares.

Figure 3: Position of the 29 samples (manually selected in different areas) measured by the spectroradiometer MINOLTA

CS1000.

difference in the total energy of the two spectral

curves being compared (viggiano, 2002), or to use

the colorimetric and spectral combined metric

(cscm) (lópez-álvarez, 2005) which is based on the

comparison of the spectra of natural illuminants

from the colorimetric and spectral points of view.

3 RECOVERY OF

REFLECTANCE SPECTRA

FROM COLOR SENSORS

In the state of the art, several solutions are proposed

to recover reflectance spectra from color values.

Some solutions are based on 3D interpolation, 2D

interpolation, non-negative matrix transformation or

hybrid methods (Kim, 2012). In (Zuffi, 2008) Zuffi

et al. demonstrated that the estimation of the spectra

of objects surfaces could be performed from data

having lower dimensionality (e.g. from a RGB

camera or from a generic N-filters camera

responses). They proposed a solution, based on a

linear model, which exploits the smoothness of

surface reflectance functions. Several other methods,

which also exploit the smoothness of surface

reflectance functions, have been proposed to

generate physically realizable natural spectra, while

some additional limitations would constrain the

recovered spectra (Peyvandi, 2011).

Several global optimization methods have been

proposed to recover the spectra reflectance (Dupond,

2002). Recent works have improved the spectra

recovery accuracy using local methods. In (Bianco,

2010) Bianco proposed a local optimization-based

method which is able to recover the reflectance

spectra with the desired tristimulus values, choosing

the metamer with the most similar shape to the

RGB-D-λ: 3D Multispectral Acquisition with Stereo RGB Cameras

691

Figure 4: Recovery of the spectral distribution of the 29 samples selected in the scene considered.

reflectances available in a training set. The

reflectance spectra of the Munsell Atlas, the Vhrel

dataset, the Gretag MacBeth Color Checker CC, and

the Gretag MacBeth Color Checker DC were used as

samples in (Bianco, 2010).

In this paper we propose to recover reflectance

spectra, from N optical filters put in front of a RGB

camera sensor, using a two-step process. In our

experiment we performed several tests in two

different setups: i) using one large-band color filter

and two color camera sensors (to simulate a 6 bands

camera sensor, see example shown in Figure 1); ii)

30 narrow-band color filters and one monochromatic

camera (see example shown in Figure 7). The idea is

to combine the global optimization method proposed

by Zuffi et al. (Zuffi, 2008) with the local

optimization method proposed by Bianco (Bianco,

2010). Meanwhile the first method is used to

initialize the recovery of the spectra reflectance, the

second one is used in a second step to refine the

recovery.

The method of Zuffi et al. (Zuffi, 2008) has been

chosen as specifically designed to address the

problem that reflectance spectra can be unfeasible:

in fact the recovered reflectance can often present

lobes outside the range [0%, 100%], or can present

atypical spikes due to the reconstruction method or

to the clipping of the recovered reflectance in the

admissible range. The algorithm (Zuffi, 2008) starts

with estimating the recovered reflectance through a

linear combination of a spectral basis and then

decomposing it into the parts of spectrum inside,

above and under the admissible range, respectively.

An iterative procedure then starts to reduce the

magnitudes of the parts above and below the

admissible range. At each step, a smoothing

procedure can be run on them. The procedure ends

when the required accuracy or the maximum

number of iterations has been reached.

The final recovery is performed using the

method by Bianco (Bianco, 2010), using the

previous one as initialization. The method (Bianco,

2010) has the advantage of using multiple spectral

bases for the reconstruction. The basis is

automatically chosen on the basis of the position in

the color space of the sample to be recovered. The

main steps of the method are reported in Figure 8,

where V

and B are fixed basis, V

is the adaptive

one, GFC is the Goodness of Fit Coefficient

(Hernàndez-Andrés, 2001), and p is a value to be

heuristically set.

The results shown in Figures 4 and 5

demonstrate that this solution is efficient. However,

some errors are noticeable. For instance in Figure 4

for samples #3 (red sample on the ladybird at right),

#4 (light pink sample on the elephant), #5 (pink

sample on the elephant), #11 (blue sample on the

wheel below the elephant) and #12 (pink sample on

the wheel below the elephant). This is mainly due to

the non-uniformity of these samples (reflective

surfaces) and of their size (which was for some of

them lower that the viewing field of the spectro-

radiometer used). The Figure 6 shows the result of

the 3D reconstruction obtained from the 6-bands

camera sensor setup considered (from 8 viewing

positions). We also performed 3D reconstruction

RGB-SpectralImaging 2016 - Special Session on RBG and Spectral Imaging for Civil/Survey Engineering, Cultural, Environmental,

Industrial Applications

692

Figure 5: Computation errors between measured values (ground thruth values measured by the spectroradiometer) and

estimated values. Depending of the object/surface properties (e.g. specular reflectance) and of the SPD of the light sources

used, some ΔE94 values are quite low (e.g. the red sample below the ladybird at lelf, sample 21 or the brown nose of the

dog, sample 28) meanwhile for other objects/surfaces (e.g. the yellow sample below the head of the ladybird at right,

sample 2 or the blue sample of the elephant, sample 9) these values are higher.

(a) (b)

Figure 6: Examples of 3D reconstruction of the scene considered viewed under illuimant D65 (a), or under illuminant F2

(b). White areas (i.e. holes) are due to occlusions between 3D objects and the background.

from the 30-bands camera sensor setup considered.

In both cases results were satisfying.

We also performed 2D image reconstruction

from the two setups considered (see Table 1 and

Figure 7) in order to compare the accuracy of the

two multi-spectral systems considered with the other

multi-spectral systems used by other research

laboratories having performed measurements for the

COSCH WG1 on the “Round Robin Test”. The

results of these comparisons is under process. In

both cases the acquisitions taken for the different

spectral bands have been automatically aligned

following the method described in (Bianco &

Cusano, 2011).

4 CONCLUSION

In this paper we introduced some key questions

concerning the acquisition of accurate color images

in cultural heritage applications inside the COSCH

project. We summarized the main criteria defined by

COSCH and used by the authors to help cultural

heritage users to have a deepen knowledge and

understanding of the constraints, preconditions and

RGB-D-λ: 3D Multispectral Acquisition with Stereo RGB Cameras

693

(a) (b) (c) (d) (e)

Figure 7: Example of a 2D scene (COSCH robin test ICON) digitized by : (a) a RGB CANON camera, (b) another RGB

CANON camera sensor with a red filter (large band filter) put in front of the camera, (c) to (e) a monochrome camera with

respectively a 400 nm, 450 nm and 500 nm filter (bandwith 10 nm) put in front of the camera. During our experiments we

used: 30 optical filters (from 400 nm to 70 nm) of 10 nm bandwith and 3 optical filters of 100 nm bandwith to compare

spectral distribution recovery form either a multispectral system based on 30 narrow-band filters or a multispectral system

based on 6 wide-band filters.

Table 1: Report form defined by COSCH. Here we reported technical data related to measurement performed on the ICON

shown in Figure 7.

Spectral imaging

device details

Spectral region Vis [400-700nm] Vis [400-700nm]

No. of bands Vis 31 Vis 5

Make and mode

The Imaging Source DMK41AU02. AS

monochrome camera

Canon 1000D

31 narrowband optical filters

coupled with a yellow wideband

filter for the second shot

Detector type CCD CMOS

Details of any optics used in the

acquisition (other than those part of

the camera)

Nikon AF Nikkor 24mm f/2.8 lens with

C-mount adapter

Canon EF-S 60mm f/2.8 Macro

lens

Capture settings (Exposure time,

averaging etc.)

Different for each filter. from 5s to 1/5

1/160s for the first shot (unfiltered),

1/60s for the second one (filtered)

Actual Spectral resolution Spectral Resol: ~ 10 nm FWHM N.A.

Details of

illuminant

Type 2x ~3200K 500-Watt 2x ~3200K 500-Watt

Details of illuminant spectrum grey body radiator grey body radiator

Position of the analyzed target

(vertical/horizontal)

45° on a stand 45° on a stand

Working distance btw target and the

system

100 cm btw target and lens 130 cm btw target and lens

Angle btw targets and lights approx 2x45°/0 approx 2x45°/0

Reconstruction matrix (Y/N) N Y

Calibration workflow (Capture to

Reflectance)

Standard linear reflectance calibration

on 99% reflectance spectralon plus

non-uniformity correction with neutral

gray card

Standard linear reflectance

calibration on 99% reflectance

spectralon plus non-uniformity

correction with neutral gray card

Description of related algorithms

reconstruction matrix computed

using the acquisition of the MCC

provided

Noise reduction methods none

2 reference image averages for the

second shot

RGB-SpectralImaging 2016 - Special Session on RBG and Spectral Imaging for Civil/Survey Engineering, Cultural, Environmental,

Industrial Applications

694

Figure 8: Main steps of the spectral reconstruction method by Bianco (Bianco, 2010).

practical aspects related to the selection and the use

of a multi-spectral acquisition system.

We also reported how color data can be estimated

from spectral data and discuss several issues related

to the measurement of reflective surfaces by 2D

imaging systems.

As future work we plan to extend the spectral

reconstruction to uncontrolled light setups, where a

light estimation and correction step is needed to

render the scene as taken under a canonical light

(Bianco, Bruna et al., 2012, Bianco, Bruna et al.,

2013, Bianco & Schettini, 2014). We will also

investigate the use of more complex metrics to

compare the acquisition taken with different devices:

instead of using single-valued metrics, we will

investigate the use of metrics able to output an error

map combining both colorimetric and structural

errors (Bianco, Ciocca et al., 2009). Often

multispectral systems are calibrated using the Gretag

MacBeth chart, which colors and substrate are not

representative of those found in art. Sometimes even

the use of an external chart is too invasive. Thus, we

will investigate the use of techniques to design ad-

hoc charts (Bianco & Schettini, 2012), even taking

the optimal samples from the artifact itself. We also

plan to investigate spatio-temporal effects, such as

those reported in the ITN-DCH project (Makantasis,

2013, Kyriakaki, 2014, Doulamis, 2015) under

uncontrolled light conditions.

REFERENCES

Bianco S., Gasparini F., Schettini R., Vanneschi L., 2008.

Polynomial modeling and optimization for

colorimetric characterization of scanners. Journal of

Electronic Imaging, Vol. 17, No. 4, pp. 043002-

043002.

Bianco S., Schettini R., Vanneschi L., 2009. Empirical

modeling for colorimetric characterization of digital

cameras. IEEE International Conference on Image

Processing (ICIP).

Bianco S., Ciocca G., Marini F., Schettini R., 2009. Image

quality assessment by preprocessing and full reference

model combination. In Proc of IS&T/SPIE Electronic

Imaging. International Society for Optics and

Photonics, pp. 72420O-72420O.

Bianco S., 2010. Reflectance spectra recovery from

tristimulus values by adaptive estimation with

metameric shape correction, J. Opt. Soc. Am. A/Vol.

27, No. 8/August, pp 1868-1877.

Bianco S., Cusano C., 2011. Color target localization

under varying illumination conditions. Proc of

Computational Color Imaging workshop, pp. 245-255.

Bianco S., Bruna A., Naccari F., Schettini R., 2011.

Towards the design of low cost colorimetric imaging

device. IEEE International Conference on Consumer

Electronics-Berlin (ICCE-Berlin), pp. 229-233.

Bianco S., Colombo A., Gasparini F., Schettini R., Zuffi

S., 2011. Applications of spectral imaging and

reproduction to cultural heritage. Digital Imaging for

Cultural Heritage Preservation, pp. 183-209.

Bianco S., Schettini R., 2011. Computational color

constancy. 3rd European Workshop on Visual

Information Processing (EUVIP), pp. 1-7.

Bianco S., Schettini R., 2012. Sampling optimization for

printer characterization by direct search. IEEE

Transactions on Image Processing, Vol. 21, No. 12,

pp. 4868-4873.

Bianco S., Bruna A., Naccari F., Schettini R., 2012. Color

space transformations for digital photography

exploiting information about the illuminant estimation

process. Journal of the Optical Society of America A,

Vol. 29, No. 3, pp. 374-384.

Bianco S., Bruna A., Naccari F., Schettini R., 2013. Color

correction pipeline optimization for digital cameras.

Journal of Electronic Imaging, Vol. 22, No. 2, pp.

023014-023014.

Bianco S., Schettini R., 2014. Error-Tolerant Color

Rendering for Digital Cameras. Journal of

RGB-D-λ: 3D Multispectral Acquisition with Stereo RGB Cameras

695

Mathematical Imaging and Vision, Vol. 50, No. 3, pp.

235-245.

Bianco S., Cusano C., Schettini R., 2015. Color Constancy

Using CNNs, IEEE Conference on Computer Vision

and Pattern Recognition Workshops (CVPRW).

Boochs F., Trémeau A., Karmacharya A., Reich T., Heinz

G., Kato Z. and Tamas L., 2015. Towards a

Knowledge Model to Document and Select

Algorithms and Processes in Cultural Heritage,

Proceedings of Taima, Hammamet, 11-16 june, pp

201-209.

De Lasarte Rigueiro Marta, 2009. Thorough

characterization and analysis of a multispectral

imaging system developed for colour measurement,

PhD Thesis of the Universitat Polytècnica de

Catalunya.

Doulamis, Anastasios, et al. 2015. Selective 4D modelling

framework for spatial-temporal land information

management system", Third International Conference

on Remote Sensing and Geoinformation of the

Environment. International Society for Optics and

Photonics.

Dupont D., 2002. Study of the reconstruction of

reflectance curves based on tristimulus values:

comparison of methods of optimization, Color Res.

Appl. 27, 88–99.

Faridul H. S., Pouli T., Chamaret C., Stauder J., Kuzovkin

D., Trémeau A. and Reinhard E., 2015. Color

Mapping: A Review of Recent Methods, Extensions,

and Applications”, Computer Graphics Forum, to be

published.

Hernàndez-Andrés Javier and Romero Javier, 2001.

Colorimetric and spectroradiometric characteristics of

narrow-field-ofview clear skylight in Granada, Spain,

Journal of Optical Society of America A, 18(2), 412–

420.

Imai F. H., 2010. Spectral reproduction from scene to

hardcopy: Part I – Multi-spectral acquisition and

spectral estimation using a Trichromatic Digital

Camera System associated with absorption filters,

Report of Munsell Color Science Laboratory,

Rochester Institute of Technology, published in

http://www.cis.rit.edu/research/mcsl2/research/reports.

php#DupontReports (last access 21/07/2015).

Kim Bog G., Han Jeong-won, and Park Soo-been, 2012.

Spectral reflectivity recovery from the tristimulus

values using a hybrid method, J. Opt. Soc. Am. A / Vol.

29, No. 12 / December, pp 2612-2621.

Kyriakaki, Georgia, et al. 2014. 4D Reconstruction of

Tangible Cultural Heritage Objects from Web-

Retrieved Images, International Journal of Heritage in

the Digital Era 3.2: 431-452.

Lahanier C., Schmitt F., 2003. Le Bœuf P. and Aitken G.,

Multi-spectral Digitisation and 3D Modelling of

Paintings and Objects for Image Content Recognition,

Image Classification and Multimedia Diffusion. An

Ontology Access to the C2RMF Database and Library

using the CIDOC-CRM, Intl. Conf. of Museum

Digitization, Antiquities, Painting and Calligraphy,

November 17-18, Taipei, Taiwan.

Liang H., Saunders D., Cupitt J., and Lahanier C., 2004.

Multispectral Imaging for Easel and Wall Paintings,

PART 6: “Examination and Documentation

Techniques” of the book Conservation of Ancient

Sites on the Silk Road, Proceedings of the Second

International Conference on the Conservation of

Grotto Sites, Mogao Grottoes, Dunhuang, People's

Republic of China, June 28–July 3, pp 267-274.

López-Álvarez Miguel A., Hernández-Andrés J., Romero

J., Campos J. and Pons A., 2009. Calibrating the

Elements of a Multispectral Imaging System, Journal

of Imaging Science and Technology, May-Jun., pp

031102-1- 031102-10.

López-Álvarez M.A., Hernández-Andrés J., Valero E.M.

and Nieves J.L., 2005. Colorimetric and Spectral

Combined Metric for the Optimization of

Multispectral Systems, Proc of the 10th Congress of

the International Colour Association, pp 1685- 1688.

Makantasis, K., Doulamis A., and Doulamis N. 2013. A

non-parametric unsupervised approach for content

based image retrieval and clustering, Proceedings of

the 4th ACM/IEEE international workshop on Analysis

and retrieval of tracked events and motion in imagery

stream. ACM.

Peyvandi S. and Amirshahi S. H., 2011. Generalized

spectral decomposition: a theory and practice to

spectral reconstruction, J. Opt. Soc. Am. A, Vol. 28,

No. 8 / August, pp 1545-1553.

Ribés A., Schmitt F., Pillay R. and Lahanier C. 2005.

Calibration and Spectral Reconstruction for

CRISATEL: An Art Painting Multispectral

Acquisition System. Journal of Imaging Science and

Technology, 49, pp. 563–573.

Saunders, D., Cupitt, J., 1993. Image processing at the

National Gallery: The VASARI Project, National

Gallery technical bulletin, 14:72.

Viggiano J. A. S., 2002. A perception-referenced method

for comparison of radiance ratio spectra and its

application as an index of metamerism, Proc. of the

9th Congress of the International Colour Association,

NY, pp. 701-704.

Zuffi S., Santini S., and Schettini R., 2008. From Color

Sensor Space to Feasible Reflectance Spectra, IEEE

Transactions on Signal Processing, Vol. 56, No. 2, pp

518-531

RGB-SpectralImaging 2016 - Special Session on RBG and Spectral Imaging for Civil/Survey Engineering, Cultural, Environmental,

Industrial Applications

696