COLOR FOR IMAGE PROCESSING
Václav Hlaváč
Czech Technical University, Faculty of Electrical Engineering
Center for Machine Perception, Prague, Czech Republic
hlavac@cmp.felk.cvut.cz
http://cmp.felk.cvut.cz/∼hlavac
Physics of color.
Humans and color perception.
Color mixing.
LECTURE PLAN
Color constancy.
Courtesy to K. Ikeuchi, T. Darrell, T. Muenzer for inspiration and some pictures found in
their teaching presentations.
1/40
COLOR IN SEVERAL DISCIPLINES
2/40
Physics.
Human vision, physiology.
Psychophysics, perception.
Computer vision.
Color several disciplines
Painting, photography, movies.
Image formation, reflection physics.
Segmentation.
Color in image analysis
Image retrieval, e.g., in image databases.
ISAAC NEWTON’s STUDIES
3/40
COLOR SPECTRUM
Color is a human interpretation of a mixture of light with different
wavelength λ (a continuum of spectral colors).
4/40
The perception of color is an entirely subjective experience.
ISAAC NEWTON’S FINDINGS
He proposed in late 1660s that white
light is a multispectral mixture. (He
thought erroneously of 7 ‘primary’
colors).
This was quite a radical idea to propose
at the time, and many philosophers, such
as Johann Wolfgang Goethe
(1749-1832), refused to believe it.
Isaac Newton (1642-1727) studied nature
of light using prism.
5/40
Newton invented the color circle to
predict how colors would look when
mixed, i.e., using a linear combination
rule.
Newton circle (1709).
Capital letters refer to
notes of the diatonic scale.
VISIBLE SPECTRUM
Retina – 4 types of
receptors.
R, G, B cones,
color vision.
6/40
Rods, monochromatic vision with
higher sensitivity.
PHYSICS OF COLOR
7/40
Spectrum of electromagnetic radiation
LIGHT AND COLOR
Light = electromagnetic radiation.
Sensors do not have direct access to color, i.e.,
wavelength λ. Exception: spectrometer.
8/40
Response of i-th sensor
Zλ2
si =
s(λ) ri(λ) dλ ,
λ1
ri(λ) is spectral density of i-th sensor,
s(λ) is spectral density of light.
where
BUNSEN PRISM MONOCHROMATOR
9/40
DIFFRACTION (GRATING)
MONOCHROMATOR
10/40
SPECTRAL REFLECTANCE OF FLOWERS
11/40
RADIOMETRY, REVIEW
E
L
V
L – radiance.
E – irradiance.
n
n – surface normal.
12/40
V – direction to observer.
L(Θi, Φi)
BRDF = f (Θi, Φi, Θe, Φe) =
E(Θe, Φe)
RADIOMETRY FOR COLOR
All definitions are now “per unit wavelength”.
All units are now per unit wavelength.
All terms are now “spectral”.
Radiance becomes spectral radiance [watts per square
meter per steradian per unit wavelength].
13/40
Irradiance becomes spectral irradiance [watts per square
meter per unit wavelength].
RADIOMETRY FOR COLOR 2
Dependence on wavelength λ is introduced into BRDF.
L becomes spectral radiance.
14/40
E becomes spectral irradiance.
L(Θi, Φi, λ)
BRDF = f (Θi, Φi, Θe, Φe, λ) =
E(Θe, Φe, λ)
In computer vision, simplified models are often used which use
relative measures instead of absolute measures.
ILLUMINATION SPECTRUM
15/40
RELATIVE REFLECTANCE
16/40
RELATIVE TRANSMITTANCE
17/40
HUMANS AND TRICHROMACY
18/40
The eye reduces all the wavelengths at a given ‘pixel’ to just the
total amount of red, green, and blue.
HUMAN EYE SENSITIVITY TO λ
CIE - Commission
Internationale de l’Eclairage
(Int. Commission on
Illumination).
Performed measurements
with many humans.
19/40
Measurements from 1931 and
improvements in later years.
COLOR METAMER
A metamer, in general, is defined as two things that are
physically different but perceived the same.
Red and green adding to produce yellow is a color metamer,
because yellow could have also been produced by a spectral
color (single light with wavelengths between green and red).
20/40
Human visual system is fooled into thinking that red+green
is the same as yellow, when in fact red+green has a
completely different spectral composition than yellow.
COLOR SPACE IN 3 DIMENSIONS
Three type of cones suggest that color is
a quantity defined in 3-dimensional
(vector) space.
21/40
How to define a color space?
Shine given wavelength λ on a screen.
User must control three pure R, G, B lights and their intensity until the
colors look identical to him.
Idea of the experimental procedure.
This is possible because of color metamers.
RGB COLOR MATCHING FUNCTIONS
Issued by CIE.
Negative lobe in red!
All wavelengths cannot be
generated with RGB
components.
22/40
Solution: convert to new
synthetic coordinate
system to make the job
easy.
XYZ COLOR MATCHING FUNCTIONS
Non-negative spectra but
primary colors X, Y, Z are
imaginary.
Y (λ) should be
coincident with the
luminance.
23/40
Normalization: equal
tri-stimulus values for an
equi-energy spectrum.
GAMUT OF PERCEIVABLE COLORS
24/40
The gamut of all perceivable
colors is a subspace of 3D
space in X, Y, Z.
Y
Color = X 0 X + Y 0 Y + Z 0 Z,
where cZ , cY and cZ are
weights in the mixture.
X
Gamut is often projected into
a plane X 0 + Y 0 + Z 0 after
normalization.
Z
GAMUT IN 2D
CIE CHROMATICITY DIAGRAM
Coordinates x, y.
X
x =
X +Y +Z
Y
y =
X0 + Y + Z
z = 1−x−y
y
0.8
All mixtures appear inside.
CIE Chromaticity Diagram 1931
530
2 degrees observer
540
510
550
0.7
560
0.6
570
λ
580
0.5
590
0.4
0.3
All visible spectral colors
appear on the outline of
the ‘horse show’.
520
25/40
600
610
630
680
490
0.2
0.1
0.0
480
470
0.1
420
0.2
0.3
0.4
0.5
0.6
0.7
x
COLOR GAMUTS OF DEVICES
26/40
y
520
0.8
CIE Chromaticity Diagram 1931
530
CRT monitor gamut
540
510
520
0.8
CIE Chromaticity Diagram 1931
y
Color printer gamut
0.8
530
540
510
550
0.7
y
560
0.6
570
λ
570
590
0.3
0.2
0.1
0.0
480
470
0.2
0.3
0.4
0.5
560
570
0.3
590
600
610
630
680
490
0.6
CRT monitor
0.7
x
0.0
λ
580
0.5
0.4
0.1
420
550
0.6
0.2
0.1
Color film gamut
540
590
600
610
630
680
490
λ
580
0.5
0.4
530
0.7
560
0.6
580
0.5
CIE Chromaticity Diagram 1931
510
550
0.7
520
0.4
0.3
600
610
630
680
490
0.2
480
0.1
470
0.1
420
0.2
0.3
0.4
0.5
printer
0.6
0.7
x
0.0
480
470
0.1
420
0.2
0.3
0.4
film
0.5
0.6
0.7
x
RGB COMPONENTS OF A COLOR IMAGE
27/40
COLOR BLINDNESS, ISHIHARA PLATES
28/40
MIXTURE OF COLORS
29/40
ADDITIVE COLOR MIXING
400 nm
500 nm
+
600 nm
500 nm
600 nm
700 nm
=
YELLOW
400 nm
500 nm
600 nm
Red plus
yellow.
Additive mixing model
holds for CRT phosphors,
multiple projectors aimed at
a screen, Polachromeslide
film, human eye cones.
700 nm
GREEN
400 nm
RED
30/40
700 nm
green
makes
SUBTRACTIVE COLOR MIXING
Applies when colors mix
by multiplying the color
spectra.
Cyan (called blue in
crayons) minus (actually
multiply) yellow makes
green.
31/40
Subtractive mixing model
holds for most photographic
films,
paint,
crayons,
printing, cascaded optical
filters.
CYAN
400 nm
500 nm
600 nm
-
700 nm
YELLOW
400 nm
500 nm
600 nm
700 nm
=
GREEN
400 nm
500 nm
600 nm
700 nm
COLOR CAMERAS
32/40
1 chip camera + filter
3 chip camera
HUE, SATURATION, VALUE
COLOR SPACE
33/40
COLOR CONSTANCY, MOTIVATION
34/40
Color constancy refers to our ability to remove the effect of the
illuminant in perceiving the color of objects.
Colorimetry versus color perception.
FAILURES IN COLOR CONSTANCY
35/40
FAILURES IN COLOR CONSTANCY (2)
36/40
FAILURES IN COLOR CONSTANCY (3)
37/40
INFLUENCE OF ILLUMINATION
38/40
INFLUENCE OF ILLUMINATION (2)
39/40
INFLUENCE OF OUTLINE
40/40
Bezold effect