Академический Документы
Профессиональный Документы
Культура Документы
DEL COLOR
2008 WORKSHOP ON
“COLORIMETRY AND COLOR IMAGING”
CIENCIA Y TECNOLOGÍA
DEL COLOR
2008 WORKSHOP ON
“COLORIMETRY AND COLOR IMAGING”
Red temática “Ciencia y Tecnología del Color”. Workshop on “Colorimetry and Color Imaging”
ISBN: 978-84-92680-76-4
INTRODUCCIÓN
Índice
7. The use of optimal stimuli in colour appearance modelling. E. Perales, C. Li, E. Chorro,
V. Viqueira, F. Martínez-Verdú, M. R. Luo ......................................................................... 37
ABSTRACT:
The main purpose of this research line is to determine a methodology in order to develop a colour
vision model with a colour difference formula. In this review, we summarize the previous
research as well as the further research and the results that we expect to find.
Firstly, we describe briefly the chosen neural model, we study the uniformity of its perceptual
space, we propose a colour difference formula for it and finally we compare the predicion’s
model with those of the other models.
Considering the results, we will propose some improvements in the model, which will be
accounted in the further research.
Keywords: Colour appearance model, colour difference formula, neural model.
[8] M. R. Luo, G. Cui, B. Rigg, “The Development of the CIE 2000 Colour-Difference Formula: CIEDE2000”,
Color Res Appl. 26, 340-350 (2001).
[9] M. R. Luo, C. Li, G. Cui, “Combining Colour Appearance model with Colour Difference Formula”, in
Proceeding of the 10th Congress of the International Colour Association (AIC’05) May 8-13, Granada,
Spain, pp. 235-240 (2005).
[10] M. R. Luo, G. Cui, C. Li, “Uniform Colour Spaces Based on CIECAM02 Colour Appearance Model”,
Color Res Appl. 31, 320-330 (2006).
[11] S. S. Guan, M. R. Luo, “A colour-difference formula for assessing large colour differences”, Color Res
Appl. 24, 344-355 (1999).
[12] A. Valberg, Light Vision Color, Editorial Wiley, England (2005).
[13] R. DeValois, K. K. DeValois, E. Switkes, L. Mahon, “Hue Scaling of Isoluminant and Cone-spedific
Lights”, Vision Res. 37, 885-897 (1997).
2. Previous Research
Our start point was the ATTD05 model, which is a
neural model that can predict the perceptual
descriptors, but initially this model is not useful for
predicting colour differences.
2.a. Description of the ATTD05 model
The ATTD05 model1-2 has three opponent
transformations, one at the retinal-LGN level and
two at the cortical level, multiplicative and
subtractive adaptation mechanisms and non-
Fig.1. Flowchart of the ATTD05 model.
linearities of the Naka-Rushton type.
Basically, the cone responses are the result of a
Thirty free parameters were determined by
first linear stage and after another lineal combination
minimizing the average distances between the
yields the responses of an achromatic channel, two
predictions of the model and a set of training
opponent channels with red-green opponency but
experimental data. The training data set included the
different polarities and one opponent channel with
spectral sensitivity of LGN opponent and non-
blue-yellow opponency. Finally the channels ATD,
opponent cells, as well as those of the chromatic and
achromatic perceptual mechanisms derived from hue experimental data set was the same data used for the
cancellation experiments in both neutral and development of CIEDE20008 and included 3657
chromatic adaptation conditions. But no data of the sample pairs with an average of 2.6 ΔE units.
colour difference were used to determine the
We found that Colour Difference Formula’s
parameters of the model.
predictions from ATTD05 were better than the
2.b. Uniformity of the ATTD05 model predictions from CIELAB for Sort Colour
Differences and better that CIELAB, CIE94 and
Next, we studied the uniformity of the
CIEDE2000 for Long Colour Difference data set,
perceptual space of the ATTD05 model4, because if
but were worse than the predictions from
the model yielded a uniform perceptual space we
CIECAM029-11.
could propose the colour difference formula as a
Euclidian distance between two points.
3. Further Research
From now on, our aim is to improve the obtained
results, and for this reason, we have proposed to
improve some stages of the model, because if the
model is better, the obtained results with it will be
better too. But restructuring the stages implies that
we have to calculate the coefficients and test the
model again. When we have a better model, we will
implement Colour Difference Formula as in the
previous way.
For this reason the future research will have to
be structured in the next way:
3.a. Restructuration of the ATTD05 model
In order to improve the model’s predictions we
propose some changes into the model, the most
Fig.2. Samples of constant Munsell Value 5 plotted on the important change is in the pre-cortical stage.
perceptual plane of the ATTD05 model.
At the first version of the model we calculated
two opponent channels with red-green opponency
In the figure 2 you can see the samples of the but different polarities because we assumed that
Munsell with a constant value in the perceptual these channels would be supported by Parvo cells
space of ATTD05. If the samples appeared with Type I receptive fields, with polarity +L-M
uniformly spaced in circles was because the mediated by L-on centre cells and polarity +M-L
perceptual space was uniform and, as you can see, mediated by M-on centre cells, and we supposed that
the space was quite uniform. But this was only a existed their symmetric receptive fields, mediated by
qualitative evaluation. Others quantitative L-off and M-off centre cells. In the new version we
parameters were calculated for study of the suppose that the responses of on-centre and off-
uniformity, as the position and eccentricity of the centre cells are not symmetric12-13, and therefore our
centre of gravity or circularity of the rings. With proposed model will have four opponent channels in
these parameters we could conclude that the this stage.
ATTD05 was uniform and therefore a colour Because the model will be restructured, the
difference could be computed as the distance first step will be to calculate the coefficients of the
between two points, with a weighing Euclidean new version of the model. Again the free parameters
metric, that is, with weight factors in each will be determined by minimizing the average
coordinate, so we had to fit experimental data set in distances between the predictions of the model and a
order to obtain the coefficients. training experimental data set. But now, the training
2.c. ATTD05 colour difference formula versus data set will include the spectral sensitivity of LGN
other models opponent and non-opponent cells, as well as those of
the chromatic and achromatic perceptual
Finally, we compared predictions4 made with mechanisms derived from hue cancellation
our colour difference formula and with other models experiments of DeValois and DeValois13.
using statistical test based on parameters as
Performance Factor5 and STRESS6-7. The
4. Expected Results
Our first change in the model has been proposed in
order to do matching the model with the visual
system and for this reason we expect that the stages
of the new model will be more similar than in earlier
version to the stages of the visual system. On the
other hand if the model has improved, the uniformity
of the model should be improved too. And therefore
the Colour Difference Formula’s predictions from
new model should be better than the predictions
from earlier version of the model.
Acknowledgements
This research was supported by the Spanish Ministry
for Education and Science by means of the grant
number DPI2005-08999-C02-02.
ABSTRACT:
Today, CCD and CMOS detector arrays present excellent features in imaging systems. To
investigate the suitability of each technology according to the application, in this work, we have
used the speckle method to determine the modulation transfer function (MTF) at different
wavelengths of the visible spectrum, for the detectors of two scientific cameras (a CCD and a
CMOS). For the CMOS detector, the differences between the MTF curves intensify as the spatial
frequency augments while the MTF decreases as the wavelength increases. For the CCD detector,
the MTF spectral behaviour does not show this trend and the differences between the MTF curves
corresponding to extreme wavelengths are not expected to be significant.
Keywords: CCD and CMOS detectors, spectral image quality, modulation transfer function
(MTF), MTF spectral variation, speckle method.
[12]. N. Guérineau, J. Primot, M. Tauvy and M. Caes, “Modulation transfer function measurement of an infrared
focal plane array by use of the self-imaging property of a canted periodic target”, Appl. Opt., 38, 631-637
(1999).
[13]. M. Marchywka and D. G. Socker, “Modulation transfer function measurement techniques for small-pixel
detectors”, Appl. Opt., 31, 7198-7213 (1992).
[14] J. E. Greivenkamp and A. E. Lowman, “Modulation transfer function measurements of sparse-array sensors
using a self-calibrating fringe pattern”, Appl. Opt., 33, 5029-5036 (1994).
[15] A. M. Pozo, A. Ferrero, M. Rubiño, J. Campos and A. Pons, “Improvements for determining the
modulation transfer function of charge-coupled devices by the speckle method”, Opt. Express, 14, 5928-
5936 (2006).
[16] G. D. Boreman and E. L. Dereniak, “Method for measuring modulation transfer function of charge-coupled
devices using laser speckle”, Opt. Eng., 25, 148-150 (1986).
[17] A. D. Ducharme, "Microlens diffusers for efficient laser speckle generation", Opt. Express, 15, 14573-
14579 (2007).
[18] G. D. Boreman, Y. Sun and A. B. James, “Generation of laser speckle with an integrating sphere”, Opt.
Eng., 29, 339-342 (1990).
[19] M. Sensiper, G. D. Boreman, A. D. Ducharme and D. R. Snyder, “Modulation transfer function testing of
detector arrays using narrow-band laser speckle”, Opt. Eng., 32, 395-400 (1993).
[20] A. M. Pozo and M. Rubiño, “Comparative analysis of techniques for measuring the modulation transfer
functions of charge-coupled devices based on the generation of laser speckle”, Appl. Opt., 44, 1543-1547
(2005).
[21] J. W. Goodman, Laser Spekle and Related Phenomena, J. C. Dainty, Ed., Vol. 9 of Topics in Applied
Physics Springer-Verlag, Berlin, pp. 35-40 (1984).
[22] L. I. Goldfischer, “Autocorrelation function and power spectral density of laser-produced speckle patterns”,
J. Opt. Soc. Am., 55, 247-253 (1965).
[23] G. D. Boreman, “Fourier spectrum techniques for characterization of spatial noise in imaging arrays”, Opt.
Eng., 26, 985-991 (1987).
[24] A. Fernández-Oliveras, A. M. Pozo and M. Rubiño, “Análisis de la variación espectral de la MTF de
videocámaras CCD mediante patrones de moteado láser”, Opt. Pura Apl., 41, 221-225 (2008). Article in
press.
[25] A. Fernández-Oliveras, A. M. Pozo and M. Rubiño, MTF spectral-variation comparison of detector arrays
used in multispectral imaging systems by speckle patterns, in Proceeding of the IS&T’s 4th European
Conference on Colour in Graphics, Imaging, and Vision. 10th International Symposium on Multispectral
Colour Science, IS&T, Springfield, VA, pp. 461-466 (2008).
[26] G. C. Holst, CCD Arrays, Cameras and Displays, JCD Publishing, Winter Park, FL. and SPIE Optical
Engineering Press, Bellingham, WA, (1996).
[27] I. Shcherback and O. Yadid-Pecht, CMOS Imagers: From Phototransduction to Image Processing, O.
Yadid-Pecht and R. Etienne Cummings, Eds., Kluwer Academic, Norwell, MA, pp. 53-74 (2004).
For measuring the MTF of solid-state cameras, PSDoutput (ξ ,η ) = [MTF (ξ ,η )]2 PSDinput (ξ ,η ) (1)
the literature cites different methods that differ
essentially in the type of target or pattern used as the where ξ and η are the spatial frequencies
object. Thus, for example, methods use bar targets9, corresponding to the horizontal and vertical
random targets10,11, canted self-imaging targets12, directions x and y, respectively.
interferometric fringes13,14. PSDoutput is determined from the speckle pattern
One of the methods to measure the MTF, captured with the detector, being proportional to the
established in our laboratory, is based on using a squared magnitude of the Fourier transform of this
laser speckle pattern as the object15-20. This method speckle pattern. In the case of a rectangular single-
is suitable for analysing the detector independently slit, PSDinput is given by21,22:
of the camera lens, given that it does not require a
PSDinput ( ξ ,η ) = I
2⎡
⎢δ ( ξ ,η ) +
(λz )2 tri⎛⎜ λz ξ ⎞⎟tri⎛⎜ λz η ⎞⎟⎤⎥ (2)
lens to project the pattern. Furthermore, using a ⎜l ⎟ ⎜l ⎟
⎢⎣ l1l2 ⎝ 1 ⎠ ⎝ 2 ⎠⎥⎦
tunable laser source, we can characterize the device
at different wavelengths, this proving indispensable where| tri( X ) = 1 − X for X ≤ 1 and zero elsewhere;
< I > 2 is the square of the average speckle
in multispectral and colour-measuring applications.
irradiance; δ(ξ,η) is a delta function; l1 and l2 are,
Speckle is an interference phenomenon that respectively, horizontal and vertical dimensions of
occurs when coherent radiation is scattered from a the single-slit; λ is the wavelength of the laser; and z
rough surface. Several techniques can be used to is the distance between the single-slit aperture and
generate the speckle pattern, such as different types the detector.
of transmissive diffusers (ground glass15, fused
silica16, microlens arrays17) or integrating spheres18- Given the geometry of the single-slit, the
20
. PSDinput can be separated into frequencies ξ and η
the horizontal PSDinput(ξ,η) is the η=0 profile of
In the former case, an aperture situated in front PSDinput(ξ,η). This means that the MTF can be
of the integrating sphere enables us to specify the determined separately for x and y directions. In the
content of spatial frequencies of the speckle pattern. present work, we determine the horizontal MTF.
Two of the apertures used to date are the single-slit18 This can be done in a similar way for the vertical
and double-slit19, both of which present advantages direction.
and drawbacks20. In this work, we have used a
single-slit situated at the exit port of an integrating
sphere. 3. Method
It bears noting that in the works cited above, 3.a. Experimental set-up
the systems analysed are based generally on
Figure 1 presents the experimental set-up used.
scientific CCD cameras, and comparisons were not
It is composed of a tunable argon-ion laser source
made between devices of different technology, nor
(130 mW) or a He-Ne laser source (λ=632.8 nm; 17
was the MTF spectral variation studied.
mW), depending on the wavelength which is being
The aim of the present work is to apply the studied, an integrating sphere to generate the speckle
optical detector-characterization method, based on pattern (inner diameter of 152.4 mm), a polarizer to
the measurement of the MTF with speckle patterns, provide a linearly polarized laser-speckle pattern, a
to the analysis at different wavelengths of the image single-slit (6 mm height and variable width), and an
quality provided by different cameras. optical bench to hold the detector, which is
connected to the control card installed in a personal
For this, we have comparatively studied the
computer.
resulting MTF curves at different wavelengths of the
visible spectrum, for the detectors of two scientific The laser radiation is aimed at the entrance port
cameras (a CCD and a CMOS). of the integrating sphere, generating the speckle
pattern at the exit port. The aperture situated at the
exit port of the sphere (single-slit) determines the
2. Theoretical background content in spatial frequency of the pattern registered
The relationship between the theoretic power- in the detector. Under these conditions, the linear
spectral density known for a single-slit (PSDinput) polarizer ensures that the PSDinput is given by Eq.
and the measured power-spectral density (PSDoutput) (2)22.
allows us to determine the MTF of the detector by
means of the expression18:
l1 Distance detector-
z= (3) Wavelength (nm)
aperture (mm)
λ ξ Ny 632.8 47
514 58
where l1 is the slit width, λ the wavelength of the 488 61
laser, and ξNy is the Nyquist spatial frequency of the 457 66
detector in the horizontal direction. For a detector
array with a centre-to-centre spacing between the
photoelements Δx, the Nyquist frequency is given 3.b. Data processing
by:
Once the detector was set at the corresponding
1 distance from the single-slit aperture, as indicated in
ξ Ny = (4) the previous section, the PSDoutput(ξ) was
2Δx
determined in the following way:
In this work, measurements were made using
For a given digitized frame of speckle data, a
the detectors of two different scientific cameras: a
region of 500x500 pixels was selected. Each
CCD and a CMOS.
horizontal row of data is a single observation of an
The scientific CCD camera had a high- ergodic random process. A fast Fourier transform
resolution CCD B/W PixelFly array of 1360x1024 (FFT), which is a discrete Fourier transform, was
pixels with a centre-to-centre spacing between them performed on each row of speckle data. The
of 4.65 µm. Consequently, with Eq. (4) taken into magnitude squared in one dimension provided a
account, the Nyquist frequency of this detector is single estimate of the one-dimensional power
107.53 cycles/mm in both directions. spectrum, PSDoutput(ξ). These 500 spectra were
averaged, for a better signal-to-noise ratio in the
The CMOS camera used was a CMOS B/W
PSDoutput(ξ)23. To reduce the noise even further, the
Atmos Areascan 1M30, the detector array of which
average was taken for 10 frames.
had 1312x1024 pixels. In this case the pixel pitch
was 5 µm in the horizontal as well as in the vertical
The frames were stored in tiff format without The MTF experimental values of the detectors
compression, using an integration time of 0.050 s for were normalized by dividing them by the same value
the scientific CCD detector and 0.004 s for the used to normalize the corresponding adjustment
CMOS detector. In the case of the video camera curve (zero-order coefficient in the polynomial-fit
CCD detector, the frames were extracted in tiff expression).
format from a video recording captured at a rate of
The results are shown in Figures 2 and 3,
25 frames/s for 1 s.
which reflect, for the different wavelengths of the
When a FFT is performed on a dataset of visible spectrum, the experimental values of the
length N, the Nyquist frequency appears at the N/2 horizontal MTF of each detector after normalization
component of the FFT output. A ratio can be formed at zero spatial frequency. For greater clarity, all the
to evaluate the spatial frequency ξn that corresponds points corresponding to the 512 MTF experimental
to the n’th component as20: values are not shown.
ξ Ny ξn
= (5)
N 2 n
Eq. (5) associates frequencies between zero and
the Nyquist frequency with FFT components from 0
to the N/2 component. In this work, we used
N=1024, thus the total number of spatial frequencies
contained in the range from 0 to the Nyquist
frequency of the detector was 512.
Before processing, each digitized frame of
speckle data was corrected in order to reduce effects
from the spatial noise of the detector itself.
With respect to the spatial noise of a CCD, a
distinction can be made between the fixed pattern Fig. 2. MTF experimental values of the scientific camera
noise (FPN) and the photoresponse non-uniformity CCD detector at different wavelengths of the visible
(PRNU). The FPN refers to the pixel-to-pixel spectrum.
variation that occurs when the array is in the dark,
and thus it is signal-independent noise. The PRNU is
due to the difference in response of each pixel to a
given signal; it is therefore signal-dependent noise.
The FPN was corrected by subtracting from the
speckle image the dark image captured obscuring the
detector, and the PRNU by means of the procedure
proposed elsewhere15.
For the processing of the speckle images, the
appropriate software was developed using
MATLAB.
4. Results and discussion Fig. 3. MTF experimental values of the scientific camera
For each wavelength analysed, the experimental CMOS detector at different wavelengths of the visible
values of the horizontal MTF of the detectors were spectrum.
calculated using Eq. (1).
For the two detectors, at each wavelength Figures 4 and 5 show, for the two detectors and
analysed, a polynomial fit of the experimental MTF at the different wavelengths, the MTF curves given
values was made and the resulting functional by the polynomial adjustments of the experimental
expression was normalized by dividing it by the values, after normalization at zero spatial frequency.
value that the adjustment equation provided at zero At the different wavelengths, MTF curves of
frequency. the scientific camera CCD detector were determined
approximately from half of the Nyquist frequency of For the scientific CCD detector, the differences
the detector). between the MTF curves corresponding to extreme
wavelengths are not likely to be significant; but they
Within the whole spectral range studied here,
are for the scientific CMOS detector.
the MTF spectral behaviour of the scientific
detectors does not show the same trend, the Since the penetration depth of photons in the
differences between the MTF curves at the different detector material increases with the wavelength,
wavelengths being more notable for the CMOS than effects of smear and signal loss are more
for the CCD scientific detector. Even the differences pronounced at higher wavelengths. Therefore, the
between the MTF curves corresponding to extreme influence of the diffusion on the MTF of the detector
wavelengths are probably not significant for the is stronger as the wavelength increases.
scientific CCD, as opposed to what appears for the
The new measurements incorporated into the
CMOS detector.
present work demonstrate this fact, as the results
Differences in the MTF spectral behaviour of found for the scientific CMOS detector illustrate. In
the detector arrays studied could be due to the effect the case of the scientific CCD detector, to observe
of charge diffusion between pixels, which depends these effects, it would be necessary to carry out new
on wavelength26. Probably, the charge diffusion measurements at higher wavelengths than those
effect is slightest for the CCD scientific camera, and analysed here.
therefore the wavelength does not influence the
In this sense, it is worthwhile to become aware
MTF of its detector significantly.
of the differences reported in the MTF spectral-
On the other hand, our results show that the behaviour of the detector arrays analysed in our
CCD detector presented MTF values higher than work, taking them into account when choosing a
those of the CMOS detector at the same spatial solid-state camera for specific applications within
frequencies, for each of the four visible wavelengths the spectral range we have studied.
analysed.
Acknowledgments
5. Conclusions The authors express their appreciation to the
In this work, we have comparatively analysed the Ministerio de Educación y Ciencia for financing
quality of the images provided by different detector projects FIS2004-06465-C02-02 and FIS2007-
arrays using the speckle method. In addition, we 66671-C02-02.
have compared their performance with wavelength
within a range of the visible spectrum. For this, we
have studied the MTF at several visible
wavelengths, for the detectors of two scientific
cameras (a CCD and a CMOS).
For all the wavelengths studied, the scientific
CCD detector presented MTF values higher than
those of the CMOS detector.
Moreover, our results reveal differences in the
MTF spectral-variation of the detector arrays
analysed within the spectral range studied.
In the case of both scientific cameras, the MTF
spectral behaviour does not show the same trend
within the spectral range studied, the differences
between the MTF curves being more notable for the
CMOS detector.
For the scientific CMOS detector, the
differences between the MTF curves arise as the
spatial frequency augments. Beyond half of the
Nyquist spatial frequency of this detector, the MTF
decreases as the wavelength increases.
ABSTRACT:
Color measurements of 18 virgin-olive-oil tasting glasses and 10 different virgin olive oils, in a
color cabinet with a D65 source. These measures were made using a spectroradiometer with
geometries tilted at 0º, 30º, and 60º, simulating different positions of the taster’s eye. None of the
cups employed had all their geometrical dimensions within the standardized values, despite being
cups used in official sensorial analyses. Comparing color variability of the oils in different tasting
cups we discovered that in all the color differences the average color difference was above visual
threshold.
Keywords: Virgin olive oil, oil tasting, color measurement, color differences, standard oil-tasting
cups, MCDM.
cup 2, from the case of oil 1 in cup 2 versus oil 2 in Table I shows the average CIELAB color difference
cup 1.
( ΔE )
*
ab i , j
and standard deviation (SD) found with
Figure 2 shows the percentage of color pairs
with CIELAB color differences in different the 13,770 color pairs possible from our 18 oil-
intervals, and in the different geometries of 0º, 30º, tasting cups and 10 virgin olive oils. In agreement
with previous results, Table 4 shows that the lowest
and 60º. For the 0º geometry the percentage of color
average color differences and standard deviations
pairs with CIELAB color differences in the interval
correspond to the 0º geometry, with values clearly
0-2 CIELAB units is close to 50%, while this
greater than the human visual threshold. In any case,
percentage is considerably lower (around 15%) for it must be said that, as expected, the use of blue-
the 30º and 60º geometries. A large percentage of tinted glasses in oil-tasting cups reduces the
small color differences (< 2 CIELAB units) can be perceived color differences between different olive
found for two tasting cups with different oils when oils: for example, in a previous work6 we found that
the visual observation is made in a horizontal when transparent Pyrex glasses with 46.4 mm
direction. This percentage of small color differences thickness were used, the average color differences
is lower than for 30º or 60º visualization. A very from the 10 oils considered here was 10.96 CIELAB
high number of pairs with clearly perceptible color units, with a standard deviation of 7.23 CIELAB
differences (> 2.0 CIELAB units): specifically, more units (clearly higher than those shown in Table 4).
than 50% of the total pairs for the 0 º geometry and
more than 85% of the total pairs for the 30 º and 60 º
geometries. There also are pairs with very strong TABLE I
color differences (>15 CIELAB units). Average CIELAB color difference and standard deviation
(SD) computed from the 13,770 color pairs obtainable
from our 18 olive-oil tasting cups and 10 virgin-olive oils,
considering each one of the 3 measurement geometries.
Angle ΔE SD
ab,10
50
0deg
30deg
0º 3,07 2,87
40 60deg
30º 5,25 3,07
30
60º 6,89 4,59
%
20
10 4. Conclusion
0
The instrumentally measured color-differences in
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 this work are greater than typical human visual
ΔE*ab,10 (CIELAB units)
thresholds in many cases. It would be reasonable in
virgin-olive-oil sensorial analyses to propose the use
Fig.2. Histogram of CIELAB color differences measured
of opaque tasting cups made with black material. In
by a combination of all potential color pairs from 18
tasting cups and 10 virgin olive oils, distinguishing the this way it would be possible to completely avoid
three measurement geometries (0º, 30, and 60º). the influence of color on the tasters’.
º
Acknowledgments
Research Project P06-AGR-01744, Consejería de
Innovación, Ciencia y Empresa, Junta de Andalucía
(Spain), with European Regional Development Fund
(ERDF) support.
Dept. de Óptica y Optometría. Universidad Politécnica de Cataluña (UPC). Centro de Desarrollo de Sensores,
Instrumentación y Sistemas (CD6). (http://www.cd6.upc.es/).
E-mail: Jorge.alexis.herrera@cd6.upc.edu
RESUMEN:
En este trabajo se hace un estudio de las herramientas multiespectrales en la aplicación de la
medición del color en iris humanos, prótesis oculares y lentes de contacto cosméticas. Con el
análisis estadístico de los valores CIELa*b*, las diferencias de color CIEDE2000 y las gamas de
color las muestran se clasifican en tres grandes grupos: marrón, azul y verdes. Así mismo, se
analiza la reproducción que de los iris humanos hacen las prótesis y las lentes de contacto. Los
resultados muestran que los iris humanos utilizados en este estudio son predominantemente
marrones y que la reproducción del color hecha por las prótesis es aceptable mientras que no es
así en el caso de las lentes de contacto. Además, se plantea como perspectiva el incluir el análisis
de no sólo los valores colorimétricos promedio, sino también el análisis de la distribución espacial
del color.
ABSTRACT:
In this work a study of multispectral tools in color measurement of irises, prostheses and cosmetic
colored contact lenses is performed. With the statistical analysis of CIELa*b* values, the
CIEDE2000 color differences and color gamuts, the samples are classified in three major groups:
brown, blue and green. In the same way, the irises color reproduction made by prostheses and
contact lenses is analyzed. The results show that brown irises are predominant into the set of
samples utilized and the irises color reproduction in prostheses is closer than in contact lenses.
Besides, we set out as perspective, to include the analysis of not only mean colorimetric values,
but also the analysis of color spatial distribution.
Keywords: Color, multi-spectral systems, iris, gamut.
[5] M. Vilaseca, R. Mercadal, J. Pujol, M. Arjona, M. de Lasarte, R. Huertas, M. Melgosa, and F. H. Imai,
“Characterization of the human iris spectral reflectance with a multispectral imaging system”, Appl. Opt.,
47, 5622-5630 (2008).
[6] J. Daughman, 1994. U.S. Patent 5,291,560. Biometric personal identification system based on iris analysis.
Issue date: March 1, 1994.
[7] P. D. Imesch, I. H. L. Wallow, and D. M. Albert, “The color of the human eye: A review of morphologic
correlates and of some conditions that affect iridial pigmentation”, Surv. Ophthalmol., 41, S117-S123
(1997).
[8] E. Pascal, E. Gooding, and L. Hannan, “Is the albino iris worse than no iris?”, Ophthalmic Physiol. Opt.,
18, 383. (1998).
[9] G. Jordan, and J. D. Mollon, “Rayleigh matches and unique green”, Vis. Res., 35, 613-620 (1995).
[10] F H. Imai, Preliminary Experiment for Spectral Reflectance Estimation of Human Iris using a Digital
Camera. Technical Report at Munsell Color Science Laboratory (2002).
(http://www.cis.rit.edu/mcsl/research/reports.php).
[11] M. Vilaseca, M. de Lasarte, J. Pujol, M. Arjona, and F. H. Imai, “Estimation Of Human Iris Spectral
Reflectance Using A Multi-Spectral Imaging System”, in Proceedings of the Third European Conference
on Colour in Graphics, Imaging and Vision (CGIV 2006), 232-236 (2006).
[12] M. Vilaseca, M. de Lasarte, J. Pujol, M. Arjona, R. Huertas, M. Melgosa and F. H. Imai, “Measuring and
Analyzing the Colour of the Iris with a Multi-Spectral Imaging System”, in Proceedings of the Fourth
European Conference on Colour in Graphics, Imaging and Vision (CGIV 2008), 427-431 (2008).
[13] M. de Lasarte, M. Vilaseca, J. Pujol, and M. Arjona, “Color measurements with colorimetric and
multispectral imaging systems”, Proc. SPIE 6062, pg. 0F1 (2006).
[14] M. Vilaseca, J. Pujol, and M. Arjona, “Spectral-reflectance reconstruction in the near-infrared region by use
of conventional charge-coupled-device camera measurements”, Appl. Opt., 42, 1788 (2003).
[15] M. Vilaseca, J. Pujol, and M. Arjona, “Illuminant influence on the reconstruction of near-infrared spectra”,
J. Imaging Sci. Technol., 48, 111 (2004).
[16] M. J. A. Port, Cosmetic and prosthetic lenses. In: Contact Lenses, 3rd ed. Eds A. J. Phillips and J. Stone,
Butterworth-Heineman Ltd, pp. 789-796 (1989).
[17] M. de Lasarte, J. Pujol, M. Arjona, and M. Vilaseca, “Optimized algorithm for the spatial non-uniformity
correction of an imaging system based on a CCD color camera”, Appl. Opt., 46, 167–174 (2007).
TABLE II
CIEDE2000 minimum color differences with respect to
real irises. Irises-Prostheses(I-P), Irises-Lenses(I-L).
∆E00 I-P ∆E00 I-L
RESUMEN:
La principal motivación de este trabajo es la búsqueda de una única expresión matemática que
permita reproducir distribuciones espectrales de forma general. Para ello se consideran
polinomios B-Spline rotacionales de segundo orden como expresión matemática base para dicha
reproducción. El objetivo fundamental de este trabajo es, por tanto, la determinación de los
coeficientes de los polinomios B-Spline que permitan reproducir distribuciones espectrales, así
como la evaluación de la exactitud de la reproducción espectral. El método utilizado para
determinar estos coeficientes es un proceso de optimización genético basado en el Algoritmo
Simplex, usando el error cuadrático medio o RMSE como función de mérito. La calidad final de
la reproducción se evalúa mediante el ‘Goodness of Fit Coefficient’ o GFC.
Palabras clave: Reproducción espectral, valores triestímulo, polinomios B-Spline.
(b)
Fig. 2. Ejemplo de reproducción de una distribución
espectral (azul) mediante un polinomio B-Spline de orden
4 (rojo) y sus puntos de control (negro). El desplazamiento
de un punto de control (a) modifica la curva sólo alrededor
del punto (b).
Fig. 1. Representación geométrica de un polinomio B-
Spline de orden 4.
En este ejemplo aparecen representadas en azul
la distribución espectral a reproducir, en rojo la
En el caso de la representación geométrica de
curva B-Spline y en negro sus coeficientes (5,
la Fig.1, la curva roja corresponde a la curva B-
polinomio de orden 4) o puntos de control. La línea
Spline descrita por un polinomio de orden 4. Los 5
negra que los une describe el lugar geométrico
puntos P0, P1, P2, P3 y P4, son la representación
donde se sitúa la curva B-Spline.
gráfica de los coeficientes o puntos de control del
polinomio B-Spline, las líneas verde, azul y rosa
3. Resultados
3.a. Antecedentes: reproducción espectral de
iluminantes
A continuación se exponen brevemente los
antecedentes a este trabajo, consistentes en la
reproducción espectral de los iluminantes A, D65, Fig. 5.(a): RMSE = 5.13338, GFC = 0.99835
F2, F7 y F11.
TABLA I
RMSE y GFC para la reproducción de las distribuciones
espectrales de los iluminantes A, D65, F2, F7 y F11 en
función del número de coeficientes del polinomio B-
Spline utilizado.
Iluminante # coefs. RMSE GFC
5 0.57182 0.99999
10 -- --
A
15 -- --
30 -- --
5 5.13338 0.99835
10 3.43171 0.99916
Fig. 5.(c): RMSE = 2.36153, GFC = 0.99965 D65
15 2.36153 0.99965
30 0.85993 0.99995
5 9.21131 0.97845
10 6.34761 0.99713
F2
15 2.61352 0.99912
30 0.95830 0.99992
5 11.4456 0.88137
10 8.42169 0.99538
F7
15 3.01522 0.99901
30 1.01877 0.99993
5 6.44981 0.91536
10 5.17431 0.99715
F11
15 2.97636 0.99911
30 0.89593 0.99991
4. Conclusiones
Se han aplicado los polinomios B-Spline en la
Fig. 6. Representación de los valores triestímulo de las 24 reproducción espectral de los valores triestímulo de
muestras de la carta CCCR obtenidos a partir de la las muestras de la carta CCCR encontrando para
reproducción espectral de las reflectancias de cada
muestras mediante polinomios B-Spline. cada una de ellas el número de coeficientes
necesario para obtener una GFC > 0.9999.
En la siguiente tabla (TABLA II) aparecen, para A partir de estos resultados se propone como
cada una de las 24 muestras de la carta CCCR, el trabajo futuro la incorporación de un algoritmo para
número de puntos de control necesarios para la variar de forma automática el número de
reproducción de su reflectancia espectral, los valores coeficientes del polinomio B-Spline, así como
triestímulo XYZ obtenidos a partir de estas también la incorporación de nuevos criterios para
reproducciones, y las diferencias de color CIELAB evaluar la calidad de la reproducción.
entre los valores triestímulo medidos y los obtenidos
a partir de la reproducción espectral con polinomios
B-Spline.
1. Grupo de Visión y Color. Universidad de Alicante. Ctra. San Vicente s/n, 03690 Alicante.
2. Grupo de Pirólisis y Combustión de Polímeros. Universidad de Alicante.
Ctra. San Vicente s/n, 03690 Alicante.
Email: veronica.marchante@ua.es
ABSTRACT:
Nanopigments are a new type of pigments. They are hybrid materials consisting of organic dyes
and layered silicate nanoparticles1. Nanopigments are already applied to make polymeric coatings
and they had shown improvement in mechanical, thermal and stability properties of the substrate
and dyes1,2.
In this work, nanopigments had been applied in a thermoplastic polymer to obtain a coloured
nanocomposite. Nanocomposite’s physical-chemical and colorimetric properties had been
assessed. However, only colorimetric characterization is going to be presented.
Keywords: Colorimetry, nanopigments, coloured polymers.
1. Introduction
Nanopigments or Planocolors® are a new kind of N
S
C H3
N
CH
C H3
3
N
S
CH3
N
C H3
CH
3
N
S C H3
N
3
CH
CH3
3
N
S
C H3
N
3
C H3
C H3
N
S
C H3
N
3
C H3
CH3
N
S H
C
3
3
N
C
N CH
N CH N C H3 N CH N CH N CH
Eindhoven (Netherland)1.
Nanopigments are hybrid materials obtained
through the combinantion of organic dye molecules Fig.1: Schematical representation of clay sheet, dye
molecule (methylene blue) and blue nanopigment.
and layered clay nanoparticles. Nanopigments gather
advantages of dyes and pigments while avoiding
their drawbacks1.
Dispersion
+ 5. Colorimetric Parameters
Colorant solution 5.a.- Spectral Reflectances
In figure 3 it is showed spectral reflectances of
Ionic Exchange pigmented polymer samples over white substrate. In
all the samples there is a peak in the interval
between 420-520 nm because all the pigments are
Washing and Filtering
blue.
Spectral Reflectance
Drying 1,0 White Substrate
0,8
Fig. 2: Scheme of nanopigments’ synthesis at laboratory. LLFTA
LLAU
0,6
LLNA (TNO)
ρW
b = (a 2 − 1) (2)
1,8 1,62
1,6
Hiding Power
1 1 − a·(ρ + ρ ) − ρ ·ρ
* * * *
(3) D (m2/L)
S= , i =W,B
i 0 ,i i 0 ,i 1,4
arc coth
b·h b·(ρ i* − ρ 0*,i ) 1,2
1,0
0,8
K = S ·(a − 1) (4) 0,6
0,4 0,23 0,24
Figures 4 and 5 show absorption and scattering 0,2
0,16
10 Transparency
1,5 9
8
T (mL/m2)
1
7 6,42
0,5 6
5 4,43
0
4
400 500 600 700 3 2,49
λ (nm) 2
Fig. 4: Absorption coefficients of pigmented polymer 1
samples. 0
LLFTA LLAU LLNA LLNA (UA)
(TNO)
Scattering Coefficient (S) Fig. 7: Transparency of coloured polymer samples for
3,5 each type of pigment.
Blue Phthalocyanine
3
Blue Ultramarine
2,5
Blue Nanopigment (TNO)
2 1,0
S
6. Conclusion
Always considering a colorimetric point of view,
from this research main conclusions we can extract
are:
- Nanopigments’ colour performance is similar or
even better than conventional pigments.
- They could be classified in a medium situation
between organic and inorganic pigments.
- They have more transparency and colouring
power than other pigments, what can be an
advantage depending on product’s final
application.
Furthermore, they can be considered an alternative
to toxic inorganic pigments because they have better
colour performance than inorganic pigments and, as
they don’t have heavy metals in their composition,
they are not contaminant
Acknowledgements
Thanks to TNO-TPD Eindhoven for allowing us to
reproduce the synthesis of Planocolors® and for
their technical and knowledge support.
Veronica Marchante would like to thank the
“Conserlleria d’Empresa, Universitat i Ciència” for
the PhD grant that she has received.
E. Perales (1), C. Li (2), E. Chorro (1), V. Viqueira (1), F. Martínez-Verdú (1), M. R. Luo (2)
ABSTRACT:
A mathematical approach is proposed for searching the chromatic adaptation matrix for the
CAT02. This approach combines the non-negativity constraint for the corresponding colours’
tristimulus values with the minimisation of the colour differences between the tristimulus values
of the corresponding colours obtained by visual observations and tristimulus values of the
corresponding colours predicted by the model, which resulted in a constrained non-linear
optimisation problem. A new matrix has been found using the MATLAB routine “fmincon”.
The performances of the CAT02 with various matrices including the original CAT02 matrix and
the new matrix are tested using the visual data sets and the optimal colours. Test results show that
the CAT02 with the new matrix is the only one combination which successfully predicts
corresponding colours for all optimal colours. Althought we have to scarify the accuracy in order
to satisfy the non-negativity constraint for the corresponding colours.
Keywords: Chromatic adaptation, colour appearance models, optimal colours.
[9] R. W. G. Hunt, C. J. Li, L. Y. Juan, and M. R. Luo, “Further Improvements to CIECAM97s”, Color Res
App., 27:164–170 (2002).
[10] CIE 15:2004, Colourimetry, 3rd ed.. CIE, Vienna (2004).
y
predicts their corresponding colours with negative 0.4
s2
tristimulus values.
0.3
On the other hand, other authors have reported
s1
problem with the chromatic adaptation transform 0.2
built in the CIECAM024,5,6. During the CIE meeting
0.1
in China in 2007, a technical committee was formed
to improve the CIECAM02. And it was found that 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
all colours which fail the CIECAM02 occur on or x
near the CIE chromaticity locus. Besides, it was
widely thought that the main problem comes from Fig.1. Chromaticity coordinates (circles) for the colours
which failed the CAT02. The full curves are the CIE
the chromatic adaptation transform, because it chromaticity locuses.
predicts the corresponding colours with negative
tristimulus values. Hence, modifications towards the
CAT02 were carried out. One of the suggestions by Therefore, in this work, we propose a general
Brill and Süsstrunk is to replace the CAT02 matrix: method for generating matrix for the chromatic
adaptation transform, which will be discussed in the
⎛ 0.7328 0.4296 − 0.1624 ⎞ next section.
⎜ ⎟
M 02 = ⎜ − 0.7036 1.6975 0.0061 ⎟ (1)
⎜ 0.003 0.0136 0.9834 ⎟⎠
⎝
by 2. Method
To modify the chromatic adaptation transform, it is
⎛ 0.7328 0.4296 − 0.1624 ⎞ necessary to know the mathematical expression for
⎜ ⎟
M BS = ⎜ − 0.7036 1.6975 0.0061⎟ (2) the chromatic adaptation transform. CAT022,3 can be
⎜ 0. ⎟
⎝ 0. 1 ⎠ compactly expressed by
Note that first two rows of the two matrices are s c = M −1ΛMs (3)
the same. The different occurs at the last row. It can
be found that for some colours the CAT02 with the where:
original matrix (eq. 1) predict corresponding colours
- s is the tristimulus values of the object colour
with negative tristimulus values, but the CAT02
under the test illuminant,
with the BS matrix (eq. 2) works fine.
- sc is the corresponding colour under the reference
The CAT02 with the MBS was also used to
illuminant.
predict the corresponding colours for optimal
colours and it also failed for many of them. Some - M is the CAT02 matrix
colours (in terms of chromaticity coordinates shown
- Λ is a diagonal matrix depending on the test and
by circles) which failed the CAT02 under different
reference illuminants and the adaptation factor and
illuminants were shown in Figure 1, where the full
is given by this expression:
curves are the CIE chromaticity locuses. It can be
⎜ 1 ⎟⎠
⎝ 0.7
0.6
⎛ R wr ⎞ 0.5
⎜β ⎟
y
⎜ Rw ⎟ 0.4
⎜ ⎟
G wr
Λ (1) = ⎜ β ⎟ 0.3
(5)
⎜ Gw ⎟
⎜ ⎟ 0.2
⎜ β
Bwr ⎟
⎜ Bw ⎟⎠
0.1
⎝
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
We can express the matrix gamma by this x
⎛ 0.38971 0.68898 − 0.07868 ⎞ overall mean colour difference. The CAT02 with
⎜ ⎟ MBS becomes slight worse compared with the
M HPE = ⎜ − 0.22981 1.1834 0.04641 ⎟ (12)
⎜ 0 ⎟ CAT02 with M02. The CAT02 with the MHPE ranks
⎝ 0 1 ⎠ the third. The CAT02 with the Mnew performs the
Firstly, the visual experimental data sets were worst, which can also be expected since more
used for testing the CAT02 performance with constraints were added when the matrix was
different matrix. The average CIELAB colour optimised.
differences are listed in Table I. There are 21 data Finally, the CAT02 with each of the four
sets from seven groups of data. The last row lists the matrices was tested using the optimal colours under
mean colour difference values of all 21 data sets. the 81 test illuminants to check the non-negativity of
the corresponding colour. CIE illuminant D65 was
TABLE I used as the reference illuminant. Under the 81
Average colour difference (CIELAB) values under each illuminants, total number of optimal colours
data set for the CAT02 with M02, MBS, MHPE, and Mnew. (TNOC) generated is 20,252,868. The test results are
listed in the Table II, where NOI means the number
Matrix for the CAT02
of illuminants under which the CAT02 predicted
Data Sets M02 MBS MHPE Mnew corresponding colours with negative tristimulus
values, and where the NOCF means the number of
csaj.da.dat 5.36 5.57 6.33 6.59
optimal colours which failed the CAT02. The
kuo.da.dat 6.94 6.83 7.94 7.36 CAT02 with the original matrix failed to predict
corresponding colours for about 13.37 percent of the
kuo.dt.dat 3.57 3.68 4.57 3.89 optimal colours. The CAT02 with MBS has a
lam.da.dat 5.61 6.06 7.18 8.06 problem for about 3.45 percent of the optimal
colours. For the CAT02 with MHPE, the CAT02
helson.da.dat 7.12 7.45 7.99 9.39 predicted negative corresponding colours for about
lutchi.da.dat 7.92 7.93 7.92 10.34 0.2 percent optimal colours. While for the CAT02
with the new matrix, it successfully predicted
lutchi.dd.dat 6.88 6.88 6.53 7.76 corresponding colours for all the optimal colours.
lutchi.dw.dat 7.46 7.85 9.14 7.63 According to the non-negativity test for the
tristimulus values of the corresponding colours, the
brene.p1.dat 7.09 7.4 6.98 10.11 best is the CAT02 with the new matrix. The results
further confirmed our mathematical approach for
brene.p2.dat 5.3 5.95 4.97 8.23
searching the chromatic adaptation matrix for the
brene.p3.dat 7.85 8.53 9.29 9.63 CAT02 is correct.
brene.p4.dat 9.99 10.83 11.74 12.6
TABLE II
brene.p6.dat 8.95 9.31 8.36 10.78 Number of illuminants (NOI) under which the CAT02
brene.p8.dat 6.41 7.6 8.86 10.29 predicted negative corresponding colours and number of
optimal colours (NOCF).
brene.p9.dat 15.51 16.19 17.44 17.79 CAT02 with different matrix
brene.p11.dat 5.29 5 3.68 7.25 Original BS HPE New
brene.p12.dat 5.64 5.76 5.04 6.43 NOI 62 61 38 0
RIT1.dat 2.92 2.93 3.48 3.8 NOCF (%) 13.37 3.45 0.2 0
RIT2.dat 5.05 5.04 5.46 5.18
RIT3.dat 4.19 4.39 6.01 5 4. Conclusions
RIT4.dat 3.38 3.26 3.69 4.15 A mathematical approach has been proposed for
searching the chromatic adaptation matrix for the
Mean 6.59 6.88 7.23 8.20
CAT02. This approach combines the non-negativity
constraint for the corresponding colours’ tristimulus
values with the minimisation of the colour
As expected, the CAT02 with its original
differences between the tristimulus values of the
matrix performs the best since the matrix was
corresponding colours obtained by visual
derived based on the 21 data sets via minimising the
observations and tristimulus values of the
Acknowledgements
This research was supported by the Ministerio de
Educación y Ciencia (Spain) under the grant
DPI2005-08999-C02-02. Esther Perales would like
to thank the Spanish Ministry for Education and
Science for the PhD grant (BES-2006-13518) that
she has received.
ABSTRACT:
In the present work it is presented a method’s combination (four-source photometric stereo and
pseudo-inverse method) that allow recovering surface information (normal vectors and albedo) of
a sample and spectral information (reflectance) in each pixel. This information is used to simulate
a set of samples under different directions of illumination in a first step, and under different
geometries of illumination and different illumination sources in a second step.
Keywords: Photometric stereo, reflectance.
[12] X. Zhang, B. A. Wandell, “A spatial extension of cielab for digital color image reproduction”, Soc. For
Info. Disp. Symp. Tech. Digest, 27, 731-734 (1996).
where q+ is the pseudo-inverse of q. If q has full images under the same directions of illumination in
rank, then q+ = (qTq)qT, where qT is the transpose of order to compare the simulated images with the
q. An estimation set of spectra S1 may then be original, 24 images under 4 different directions of
obtained from the corresponding set of camera illumination makes a total number of 96 test images.
responses q1 by applying the transformation D, that
is:
S1 = Dq1 (5) DigiLite
1
Using this method, there is no need to use any
mathematical bases or to know the spectral 0.8
normalized SPD
sensitivity of the camera sensors.
0.6
2.c.- Computations
Two different simulations have been made in 0.4
this work. The first one uses the information
recovered with the photometric stereo method in 0.2
order to simulate the sample under different
directions of illumination. Once we have recovered 0
400 450 500 550 600 650 700
albedo and normal, we can consider equation (1) wavelength (nm)
with only one direction of illumination: Fig.1. Normalized SPD of the fluorescent lamp used in the
first set of simulations.
I = ρ(L·N) (6)
That way, using just one light direction we can
Two metrics were used to check the accuracy
simulate the intensity in each point of the sample
of results: root mean square error (RMS) over RGB
under any direction of illumination.
differences between the two images, and
The second application extends the previous SCIELAB12 differences.
simulation to a more general case. It is possible to
Results over the 96 images can be seen in
model albedo in each pixel as:
Table I. The first row shows several statistics over
ρ =∑E(λ)S(λ)Q(λ) (7) RMS differences and the second row the same
statistics over SCIELAB differences. Since RGB
where E(λ) is the spectrum of the illuminant, S(λ) is
values goes from 0 up to 65536 our RMS results are
the reflectance of that pixel and Q(λ) represents the
quite good. SCIELAB differences are very low too
camera’s sensitivity for each channel. Starting on the
with the mean color differences under 5 units.
results from the photometric stereo step, it is
possible to use albedo to recover reflectance9. Once
we have reflectance in each pixel of the sample, we TABLE I
only need the spectrum of one illuminant and the Results for application I.
camera’s sensitivities in each channel to simulate the Std Perc Perc
Max Min Mean
sample not only under any direction of illumination, Dev 95 99
but under any illuminant too. RMS 315 0 27 21 66 102
SCIELAB 65.47 0.03 4.27 3.57 11.54 18.46
200
SCIELAB differences. In both histograms peaks are
1000
displaced to the right, what means that differences
0 0
0 5 10 15 20 25 30 35 40 45 50 55 60
RMS Diferences
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
SCIELab differences are higher than before, but even in this case original
Fig.2. Example of simulation I: Gris 1390 simulated and and simulated images are quite similar.
captured under 270º.
3.b. Application II
In the second set of simulations, we started
again from normals and albedo recovered from the
same set of images as before. But now, we recovered Original Simulated
reflectance from albedo for each one of the 24 1200
Alabastro 1384, 90º, Incandescent
3000
Alabastro 1384, 0º, Incand
Fig. 3. The samples were captured too under those 400 1000
the three illumination sources, so 288 images were Fig.4. Example of simulation II: Alabastro 1384 simulated
compared. and captured under 0º and incandescent lamp.
1
Digilite 4. Conclusions
0.8 Incandescent
Tri-Lite In conclusion, we have shown that four-source
normalized SPD
0.6
photometric stereo allows a simple 3D modelling of
surface properties under Lambertian assumption. We
0.4 have developed and efficient method to simulate
several real-world textured objects under different
0.2 illumination geometries. Finally it is possible to use
albedo in spectral analysis to recover spectral
0
400 450 500 550 600 650 700
reflectances from linear pseudo-inverse.
wavelength (nm)
Fig.3. Normalized SPD of the lamps used in the second set
of simulations. Acknowledgements
This work was supported by Spanish Ministry of
Same metrics as before were used to analyze results. Education and Science and FEDER through the
Global statistics can be seen in Table II. grant number FIS2007-60736.
TABLE II
Results for application II.
Std Perc Perc
Max Min Mean
Dev 95 99
RMS 9285 1 588 884 2564 4486
SCIELAB 244.44 0.05 18.96 15.60 52.19 74.15
RESUMEN:
El objetivo de este trabajo es estudiar la evolución temporal de distintas fuentes de iluminación
disponibles en cabinas de iluminación para evaluar cómo varían sus acuerdos con los iluminantes
a los que representan. Para ello se emplean parámetros como la temperatura correlacionada de
color, el índice CIE de rendimiento en color, distintas métricas espectrales y las diferencias de
color CIEDE2000 entre las muestras de la carta GretagMacbeth ColorChecker iluminadas por la
fuente simuladora o por su iluminante asociado. Se concluye que, cuando necesitamos realizar
medidas colorimétricas de alta precisión, el estado de la fuente debería ser conocido, ya que es
temporalmente dependiente y no siempre se puede asumir que sea idéntico a un iluminante CIE.
Palabras clave: Fuentes de luz, iluminantes, CCT, CRI, CSCM, SCI, RMSE, WRMSE,
CIEDE2000.
TABLA I
Métricas para el análisis de datos espectrales.
RMSE WRMSE
Desv. Est. Desv. Est.
Fuente Promedio Máx Mín Promedio Máx Mín
(CV) (CV)
SLIII 0.0029 0.1364 0.1210 0.0005 0.0135 0.0110
D65
0.1315 0.0121
(2.2%) (10h) (100h) (4.1%) (0h) (100h)
SLIII 0.0012 0.0440 0.0397 0.0001 0.0042 0.0037
A
0.0419 0.0040
(2.9%) (75h) (2h) (2.8%) (75h) (2h)
SLIII 0.0103 0.2042 0.1642 0.0005 0.0314 0.0296
TL84
0.1712 0.0306
(6.0%) (0h) (0.5h) (1.8%) (100h) (0h)
CAC 60 0.0087 0.4078 0.3738 0.0030 0.0549 0.0434
D65
0.3796 0.0450
(2.3%) (0h) (1h) (6.6%) (0h) (1h)
CAC 60 0.0037 0.0909 0.0768 0.0005 0.0122 0.0102
F
0.0841 0.0113
(4.4%) (100h) (0h) (4.8%) (100h) (0h)
CAC 60 0.0071 0.1905 0.1615 0.0016 0.0338 0.0275
TL84
0.1843 0.0325
(3.8%) (100h) (0h) (4.8%) (100h) (0h)
SCI CSCM
Desv. Est. Desv. Est.
Fuente Promedio Máx Mín Promedio Máx Mín
(CV) (CV)
SLIII 0.0016 0.0519 0.0439 7.36
D65
0.0473 8.81 0.51 (5.8%) 9.78 (0h)
(3.3%) (0h) (100h) (100h)
SLIII 0.0003 0.0101 0.0090
A
0.0095 9.34 0.40 (4.3%) 9.88 (2h) 8.51 (0h)
(3.1%) (75h) (2h)
SLIII 0.0039 0.1349 0.1220 26.16 18.84
TL84
0.1297 24.84 1.90 (7.7%)
(3.0%) (100h) (1h) (5h) (0h)
CAC 60 0.0114 0.2169 0.1728 22.42 19.40
D65
0.1795 21.16 0.82 (3.9%)
(6.4%) (0h) (1h) (100h) (0h)
CAC 60 0.0012 0.0283 0.0239 32.65 28.88
F
0.0262 30.82 1.01 (3.3%)
(4.5%) (100h) (0h) (100h) (0h)
CAC 60 0.0051 0.1434 0.1210 26.02 15.18
TL84
0.1373 24.17 2.79 (11.6%)
(3.7%) (100h) (0h) (100h) (0h)
En la tabla se observa que los resultados En la tabla III se muestran los cálculos de CRI.
obtenidos para las cuatro métricas no son iguales. Las fuentes A de la cabina Spectralight III y TL84
Esto se debe a que cada métrica está basada en de la CAC 60 son las que sufren las menores
distintas propiedades de las distribuciones variaciones en CRI con el envejecimiento. Las
espectrales de las fuentes. También se observa que variaciones de aproximadamente una unidad en este
los valores crecen ligeramente con el tiempo (esto índice pueden considerarse no significativas para
es, el acuerdo iluminante – fuente simuladora nuestros propósitos. Como las variaciones que
empeora con el tiempo), excepto para la fuente D65 hemos obtenido son de este orden o menores,
de la cabina Spectralight III, en la cual los valores de podemos concluir que no son significativas.
las métricas disminuyen con el envejecimiento de la
fuente. TABLA III
En la tabla II se muestran los resultados de los Evolución temporal del índice CIE de rendimiento en
cálculos para la CCT. Puede verse como la fuente A color.
de la cabina Spectralight III es la que tiene un menor CRI
coeficiente de variación. Por el contrario, la fuente Desv. Est.
Fuente Promedio Máx Mín
D65 de la cabina CAC 60 es la que tiene el (CV)
coeficiente de variación más grande. La fuente D65 SLIII 0.44 95
D65
95 94 (0h)
de la cabina Spectralight III es la única cuya CCT (0.5%) (20h)
aumenta con el envejecimiento de la fuente. La SLIII 0.05
A
98 98 (2h) 97 (0h)
fuente D65 sufre una gran disminución en la CCT (0.1%)
con el envejecimiento. SLIII 0.27
TL84
97 98 (0h) 96 (1h)
(0.3%)
CAC 60 0.44 94
TABLA II D65
95 96 (0h)
(0.5%) (100h)
Evolución de la temperatura correlacionada de color. CAC 60 0.44 91
F
91 92 (0h)
CCT (K) (0.5%) (100h)
Desv. CAC 60 0.09
Fuente Promedio
Est. (CV)
Máx Mín
TL84
97 97 (0h) 97 (75h)
(0.1%)
SLIII 6129 36 6263 6084
D65 (6500) (0.6%) (100h) (2h)
SLIII 2823 2833 2814 En la tabla IV se presentan las desviaciones
A
6 (0.2%) estándar de las diferencias de color entre iluminante
(2856) (0h) (2h)
SLIII 3698 30 3761 3663 y fuente simuladora a lo largo del tiempo en
TL84 (4000) (0.8%) (0.5h) (100h) unidades CIEDE2000. En la primera columna
CAC 60 6117 133 6539 5992 mostramos el promedio de estas desviaciones
D65 (6500) (2.2%) (0h) (100h) estándar para todos los chips a lo largo del tiempo
CAC 60 2549 13 2574 2527 para cada fuente. En la segunda se encuentra la
F (2856) (0.5%) (0h) (100h) máxima desviación estándar para cada fuente y el
CAC 60 3747 47 3883 3705 chip que la produce. En la tercera columna se
TL84 (4000) (1.2%) (0h) (100h) encuentra la mínima desviación estándar y su chip
asociado.
TABLA IV
Desviaciones estándar en las diferencias de color
CIEDE2000 para los chips de la carta GretagMacbeth
ColorChecker.
Chips y Desv. Est. en unidades CIEDE2000
Fuente Promedio Máx Mín
SLIII
D65
0.055 0.158 (13) 0.003 (24)
SLIII
A
0.012 0.030 (15) 0.000 (23)
SLIII
TL84
0.056 0.192 (13) 0.001 (24)
CAC 60
D65
0.091 0.409 (13) 0.002 (20)
CAC 60
F
0.039 0.118 (15) 0.002 (24)
CAC 60
TL84
0.120 0.437 (13) 0.000 (24)
4. Conclusión
Este trabajo está relacionado con el trabajo actual
del Comité Técnico de la CIE TC1-44. Se observa
que la fuente A de la cabina Spectralight III es la que
sufre las menores variaciones con el envejecimiento,
atendiendo a la mayoría de los índices. La fuente
D65 de la cabina Spectralight III se aproxima mejor
a su iluminante asociado conforme envejece. Por el
contrario, la fuente D65 de la cabina CAC 60 sufre
un envejecimiento más acusado que el resto.
Podemos concluir que, cuando queremos
realizar medidas colorimétricas de alta precisión
(como experimentos sobre umbrales visuales), es
necesario conocer el estado de la fuente de
iluminación que estamos empleando, ya que es
temporalmente dependiente y no siempre podemos
asumir que es idéntica a un iluminante CIE.
Agradecimientos
Proyecto de Investigación FIS2007-64266,
Ministerio de Educación y Ciencia, España.