Light and Color
CS148, Summer 2010
Siddhartha Chaudhuri
1
Rainbow over the Potala Palace, Lhasa, Tibet (Galen Rowell, 1981)
What is Light?
Characteristics of Light
Electromagnetic radiation
Light usually refers to radiation visible to human eye
Exhibits both wave and particle-like characteristics
Has wavelength (), frequency (), amplitude etc.
Emitted in discrete quanta, called photons
Travels through vacuum at c = 299,792,458 m/s
Travels through other media at lower speeds
(Wikimedia Commons)
electron
electron
photon
positron
positron
Reconciled by quantum electrodynamics
(Image: Wikimedia Commons)
Radiometry
Light carries radiant energy
hc
A light source emits photons over time. Its power
(energy per unit time) is called the radiant flux .
Some Spectral Radiant Flux Curves
Energy of a single photon = h =
Outdoor daylight
Incandescent bulb
Mercury lamp
SP65 triphosphor fluorescent
Let radiant flux of photons with wavelengths in
range [ + ) be
The spectral radiant flux at wavelength is
Modeling radiant flux as a continuous d
function, this becomes the derivative
d
General Electric Co., 2010
Why do I need to know all this?
To produce photorealistic images of virtual 3D
scenes, by simulating actual light transport
(we won't study this too much in this course, take CS348b if you
want an in-depth treatment)
8
Just to convince you...
Photometry
Arnold van Duursen, rendered with fryrender
Perceptual study of light
Brightness, color etc. depend on the interaction of
our eyes with light
Rod (brown) and cone (green) cells
( Visuals Unlimited, 2009)
Photopic Luminous Efficiency
10
Tristimulus Theory
Cone cells perceive color (less sensitive than rods)
Three types of cone cells: (L)ong, (M)edium and
(S)hort
Responsivity
0.5
Overall sensitivity of human eye to different parts of the spectrum
(in bright lighting conditions)
Wavelength (nm)
11
Normalized response curves of human cone cells
12
Tristimulus Theory
Color Matching Experiments
Presence of 3 cone types suggests: 3 parameters
describe all colors
Two light sources with different spectral
distributions can appear to be the same color.
Such pairs are called metamers.
Wright and Guild (1920s)
Primaries standardized by the CIE in 1931
Different spectra can appear the same color (Hughes, Bell and Doppelt)
Choose lights of 3 different primary colors
Show a user single-wavelength light
Ask her/him to match it with a weighted combination
of the primaries
Red (R): 700 nm
Green (G): 546.1 nm
Blue (B): 435.8 nm
13
14
Examples
Grassmann's Law
=
=
Chromatic sensation is linear
Let:
Beam 1 color-match (R1, G1, B1)
Beam 2 color-match (R2, G2, B2)
Then:
Beam 1 + Beam 2 matches
(R1 + R2, G1 + G2, B1 + B2)
Holds for any set of primaries of any size, defines
additive color model
15
Color Matching Functions for CIE RGB
16
What's a negative amount of color???
Amounts of the red, green and blue primaries needed to match any color 17
Mathematical convenience, doesn't really exist
Implies that experimenters had to add the
corresponding amount of primary to the source
beam to get the colors to match
It turns out there is no small set of physically
realizable primaries such that any visible color is
formed by combining them with positive weights
And computing with negative numbers was tricky
in 1931
18
Enter CIE XYZ...
Oops.
Luckily, the CIE has a solution...
Let's have imaginary primaries (oh great...)
Construct linear, possibly non-realizable,
combinations of RGB primaries s.t. color matching
functions are positive throughout visible range
We have a great deal of flexibility in this choice
We want one color matching function to approximate
photopic luminous efficiency as closely as possible
Used to separate relative luminance (loosely, brightness)
from chromaticity (loosely, hue)
19
The CIE XYZ Primaries, 1931
Color Matching Functions for CIE XYZ
Map RR + GG + BB to XX + YY + ZZ
Conversion between XYZ and RGB coordinates
[] []
[
X
Y = M
Z
R
G
B
[]
[]
R
X
1
G = M
Y
B
Z
0.49
0.31
0.20
M = 0.17697 0.81240 0.01063
0.00
0.01
0.99
Note that M 1 has negative elements XYZ basis is
not physically realizable
21
xyY: Separate Chrom from Lum
X
X Y Z
y =
Amounts of the XYZ primaries needed to match any color
(y function is precisely CIE-standardized photopic luminous efficiency, 1931) 22
CIE XYZ and RGB Chromaticity Gamuts
CIE Y
Formed nonlinearly from X, Y and Z
x =
20
Y
X Y Z
Spectral colors (single-wavelength)
lie on this curved boundary
CIE Green
x and y are chromaticity coordinates, Y is relative
luminance coordinate
Gamut: Set of colors that can be physically realized
in a color space (without negative coefficients etc.)
This part of the gamut
(outside the sRGB
triangle) cannot be
displayed on normal
monitors. It's shown
here using the closest
displayable colors
sRGB
y
CIE Red
Nonspectral purples
CIE Z
23
CIE Blue
All visible chromaticities mapped to xy plane
CIE X
24
Some More Color Representations
Remember: Gamuts are actually 3D!
sRGB: HDTV/monitor/digicam standard color
space, similar to but smaller gamut than CIE RGB
Perceptually uniform representations: Distance
between two colors in color model reflects ability
of humans to discriminate between them
HSL color model
HSV color model
L*a*b* color space
(SharkD@Wikimedia Commons)
Colors projecting inside 2D sRGB chromaticity gamut may not be within true 3D gamut
(Lindbloom, 2007)
25
Reflected Light (Pigments)
26
CMYK Separation is Not Unique
Subtractive color model: color of pigment is defined
by light it does not absorb
Full black
Different components of white
light are absorbed by pigments
(C)yan, (M)agenta, (Y)ellow, (K)ey
black: typical printer primaries
27
(Images: Wikimedia Commons)
Raster Images
Raster Images
Array of pixels (picture elements)
We'll consider only 2D arrays
Binary: 0/1 per pixel
Greyscale: Single value (channel) per
pixel, usually floating point number
in [0, 1] or byte in [0, 255]
0: black, 1: white, intermediate value: grey
Interpretation depends on color space
Value of channel = weight of component
640x480, 1024x768 etc.
Color/Bit Depth: Number of bits per pixel
Color: Multiple channels per pixel,
usually RGB triplet
Resolution: Number of pixels
Types of images
28
No black
24-bit color image has 8 bits (1 byte) per channel
Transparency
Alpha channel added to pixel (e.g. RGBA)
= 0 pixel is fully transparent
= 1 pixel is fully opaque
=0
29
=1
30
