Presentación de la Universidad de Granada sobre cambios de color en escenarios naturales debidos a la interacción entre luz y atmósfera, realizada durante las jornadas HOIP 2010 organizadas por la Unidad de Sistemas de Información e Interacción TECNALIA.
Más información en http://www.tecnalia.com/es/ict-european-software-institute/index.htm
2. COLOR CHANGES IN A NATURAL SCENE
Javier Romero DUE TO THE INTERACTION BETWEEN THE
Professor
LIGHT AND THE ATMOSPHERE
Juan L. Nieves
Associate Professor
• Motivation and State of the Art
• Physical model
• Experiment
Javier Hernández-Andrés
Associate Professor
• Colour changes with distance
• Conclusions and future work
Raúl Luzón
Ph.D. student
4. Motivation
• Single Scattering : ( Mie 1908 )
Light is degraded due Incident
to its interaction with
Beam
molecules and particles Size: 0.01 μm Size: 0.1 μm Size: 1 μm
in the atmosphere.
• Multiple Scattering :
Degradation depends First Order
Third Order
on the range (distance)
Incident Beam
and on the wavelength.
Second Order
5. Motivation
Light is degraded due
to its interaction with
molecules and particles
in the atmosphere.
* reduction in visibility and contrast
* color changes:
-less saturated colors,
-hue change,
Reversibility?
6. Motivation
Color, size, shape,
texture are the main
Foggy Day Image
features for pattern
recognition...
...in addition to spectral Clear Day Image
information which can
influence surveillance Are color and spectral
and identification. degradation reversible?
“De-weathering” images?
7. State of the Art
Current image enhancement algorithms
1) Non-physics-based algorithms:
• Based on statistical information of the image,
• ... using no information about the imaging physics.
2) Physics-based models:
• Using the underlying physics of the atmospheric
degradation process...
• ...and then to compensate for it with appropriate
image processing.
8. State of the Art
1) Based on statistical information of the scene:
Histogram equalization and its variations (Pitas and
Kiniklis [1996], Pizer et al. [1987]).
•RGB channels as separate channels
•Certain improvement on HSI space.
Advantages Drawbacks
Straightforward technique False colors
Not intensive computation Undesirable effects
Increase the global contrast
Histogram
Original equalized
9. State of the Art
2) Physics-based models:
Light interaction with particles and molecules of
different sizes in the atmosphere:
• Absorption-Emission;
• Scattering:-Attenuation
-Airlight
McCartney
[1976]
10. State of the Art
2) Physics-based models:
The best physical based models are those constructed over the
dichromatic atmospheric scattering model (Tan and Oakley
[2001], Narasimhan and Nayar [2000]).
These models are based on
single-scattering.
Assuming the same β for all color
channels…
Narasimhan and Nayar (2003) …the color of a scene point is a
linear combination of the direction of
airlight and the direction of direct
transmission (attenuated by
scattering)
11. State of the Art
2) Physics-based models:
Advantages Drawbacks
Exploit the underlying Usually needs information
physics of the degradation about meteorological
process conditions
Good color recuperation Some images taken under
different weather conditions
Applicable for different Identify some points on the
distances scene
Simplification of real process
12. State of the Art
Original Enhanced with physical model
RGB HSI
From Tan and Oakley [2001]
13. Our goal
Simple and fast algorithm to recover color
information (and spectral information)... for clear
days and overcast days.
Only one image: no distance information, and no
scattering coefficients values.
But, we need first to analyze and to quantify the
color changes due to the atmosphere.
14. • Motivation and State of the Art
• Physical model
• Experiment
• Colour changes with distance
• Conclusions and future work
15. Physical Model
The irradiance (E) in one pixel is proportional to the radiance
of the scene (L), assuming there is no absorption and
λ
reflection inside the camera E ( ) = ΩL( ) λ
For perfect Lambertian
surfaces
ρ ( λ ) Ed ( λ )
LO (λ ) =
π
16. Physical Model
Radiance from the object at the camera plane has two terms
(Narasimhan and Nayar [2000], [2003]):
• one due to direct light coming from the object and
attenuated by the atmosphere
• other term: airlight
− βtot ( λ ) d − βtot ( λ ) d
L(λ ) = L0 (λ )e + L∞ (λ )(1 − e )
Direct light Airlight
where:
L is the object radiance viewed from the observer plane
L0 is the object radiance
βtot = βsct + βabs , is the attenuation coefficient in the
atmosphere
L∞ is the radiance of the horizon
d is the distance between the object and the detector
λ is the wavelength
17. Physical Model
For clear skies, a Lambertian object receiving an
irradiance Ed produces an irradiance on the detector :
Ed (λ ) ρ (λ ) − βtot ( λ ) d − βtot ( λ ) d
Et (λ ) = Ω e + ΩL∞ (λ )(1 − e )
π
where:
Ω is the solid angle subtended from the object into the
detector
Ed is the irradiance over the object
ρ is the spectral reflectance of the object
βtot is the attenuation coefficient
d is the distance between the object and the detector
L∞ is the horizon radiance
λ is the wavelength
18. Physical Model
For overcast skies, assuming an homogeneous
distribution of the sky radiance [Gordon and Church
[1966]) and a Lambertian object:
Et ( λ) =ΩL∞ ( λ) ρ ( λ) e
−βtot ( λ) d
(
+ΩL∞ ( λ) 1− e
−βtot ( λ) d
)
where:
Ω is the solid angle subtended from the object into the
detector
ρ is the spectral reflectance of the object
βtot is the attenuation coefficient
d is the distance between the object and the detector
L∞ is the horizon radiance
λ is the wavelength
19. • Motivation and State of the Art
• Physical model
• Experiment
• Colour changes with distance
• Conclusions and future work
20. Experiment
Color changes
CIE 1931 (x,y,Y) and CIELAB (L*,a*,b*) values
corresponding to 240 objects of the GretagMacbeth
Color-Checker DC, whose spectral reflectances are
known
GretagMacbeth SpectraScan PR-650
ColorChecker DC spectroradiometer
23. Experiment
Ed (λ ) ρ (λ ) − βtot ( λ ) d − βtot ( λ ) d
Et (λ ) = Ω e + ΩL∞ (λ )(1 − e )
π
We know the scattering coefficient at 450, 550 and 700
nm and we can interpolate to the rest of visible spectrum
assuming that (McCartney [1976]):
1
β sct = cte
λ u
Another assumption: the absorption coefficient is constant
in the visible range.
29. Colour changes in the object
with observation distance
Direct light from the
object is attenuated
with the distance
For a specific
distance, airlight
becomes more
important.
33. Colour changes in the object
with observation distance
CIELAB
Are these colour changes reversible?
Are we able to enhance visibility for
better identification?
…if so, some kind of colour constancy
could be achieved.
34. ...and what does “color constancy”
mean?
Colour appearance can chage dramatically under
different illumination conditions…
…finding both a color mapping and the color of the scene
illuminant are equivalent problems.
35. ...and what does “color constancy”
mean?
Colour appearance can chage dramatically under
different illumination conditions…
CCT = 2760K
Incandescent
lamp
CCT = 5190K
Day-light
…but the human visual system is able to
☺ compensate for those chages.
36. ...and what does “color constancy”
mean?
Cones excitations change
regularly with illumination
What about the images degradated by the atmosphere?
37. ...and what does “color constancy”
mean?
For a particular object:
L viewed under different distances
versus
L under the E illuminant (flat spectrum)
Same for M and S cones or for just R, G, B
Ed (λ ) ρ (λ ) − βtot ( λ ) d − βtot ( λ ) d
Et (λ ) = Ω e + ΩL∞ (λ )(1 − e )
π
Clear days
Et ( λ) =ΩL∞ ( λ) ρ ( λ) e
Overcast days
−βtot ( λ) d
(
+ΩL∞ ( λ) 1− e
−βtot ( λ) d
)
38. ...and what does “color constancy”
mean?
20 objects from the Color Checker
For a zero distance we should expect a
linear relation:
Other distances?
L
Other cones (M or S)?
Other broad band sensors
(R,G,B)?
LE
39. ...and what does “color constancy”
mean?
20 objects from the Color Checker
For a zero distance we should expect a
linear relation:
Other distances?
Other cones (M or S)?
Other broad band sensors
(R,G,B)?
40. Conclusions and future work
It`s clear that visibility of objects depends on weather
conditions and changes in the objects’ color can
influence identification.
Colour constancy
approaches could be ?
applied in bad weather
conditions to restore
the colour appearance
of objects.
41. Javier Romero
Professor
Thank you for your
Juan L. Nieves
attention!
Associate Professor
Javier Hernández-Andrés
Associate Professor
Raúl Luzón
Ph.D. student
42. References
1. W. E. K. Middleton, “Vision through the atmosphere”, 2nd Edition, University of Toronto Press, 1952
2. I. Pitas and P. Kiniklis, “Multichannel Techniques in Color Image Enhancement and Modeling”, Image
Processing, IEEE Transactions, Vol 5,No. 1, pp. 168-171, 1996.
3. Stephen M. Pizer, E. Philip Amburn, John D. Austin, Robert Cromartie, Ari Geselowitz, Trey Greer,
Bart ter Haar Romeny, John B. Zimmerman and Karel Zuiderveld, “Adaptive histogram equalization and
its variations”, Computer Vision, Graphics and Image Processing Vol 39, 355-368, 1987.
4. K. Tan and J.P. Oakley, “Physics-Based Approach to Color Image Enhancement in Poor Visibility
Conditions”, Journal of the Optical Society of America, Vol. 18, No. 10, pp. 2460-2467, 2001.
5. S. G. Narasimhan and S. K. Nayar, “Chromatic Framework for Vision in Bad Weather”, Conference on
Computer Vision and Pattern Recognition, IEEE Proceedings. Vol. 1, pp. 598-605, 2000.
6. S. G. Narasimhan and S. K. Nayar, “Contrast Restoration of Weather Degraded Images”, Pattern
Analysis And Machine Intelligence, IEEE Transactions, Vol. 25, No. 6, pp. 713-724, 2003.
7. S. G. Narasimhan and S. K. Nayar, “Vision in Bad Weather”, Seventh IEEE International Conference in
Computer Vision, IEEE Proceedings, Vol 1, pp. 820-827, 2000.
8. Earl J. McCartney, “Optics of the atmosphere, scattering by molecules and particles”, Wiley-
Interscience, 1976.
9. Nascimento SMC, Ferreira FP, Foster DH. “Statistic of spatial cone excitation ratios in natural scenes.
J Opt Soc Am A ;19:1484–1490 (2002).