CS 450: Introduction to Digital
Signal and Image Processing
Color Processing
Light and Wavelengths
• Visible light is in the range 400 nm (blue)
to 700 nm (red).
Perception of Light
• Cones have three different kinds of color-sensitive
pigments, each responding to a different range of
wavelengths.
• These are roughly “red”, “green”, and “blue” in their
peak response but each responds to a wide range of
wavelengths.
• The combination of the responses of these different
receptors gives us our color perception.
• This is called the tristimulus model of color vision.
Cones and Wavelengths
(From G&W, individual cone responses normalized)
Cones and Wavelengths
• The sensitivity and
number of the three
types of cones are
different
• More sensitive
overall to green and
red than to blue
Relative responses of the three different cones
Luminous Efficiency Function
• The luminous efficiency function combines the
responses of all three to measure perceived
brightness for different wavelengths
RGB Color Model
• Simplest model is just to store red, green,
and blue values
• Colors can be thought of as points in a
RGB cube
RGB Color Channels
“Web Safe” Colors
Primaries and Secondaries
• Primary colors: ones
mixed to make other
colors
• Secondary colors:
pairwise
combinations of
primaries
• Can be additive or
subtractive
CMY Model
• Subtractive media absorb rather than emit light
– Real-world objects
– Paint
– Printing ink
• The perceived color is what is not absorbed
(reflected)
• CMY based on subtractive primaries:
cyan (C), magenta (M), and yellow (Y)
CMYK Model
• Problem with CMY model:
– Because of imperfect primaries, can’t entirely
absorb all colored light (i.e., make black)
– Usually produces a dark greyish brown
• Solution: add black (K) as a fourth primary
• CMYK most common model for printers
Luminance and Chromaticity
• RGB and CMYK aren’t the most intuitive
model of colors.
• Artists usually think of dark/light and color
as two different things:
– Luminance (brightness)
– Chromaticity (color)
Luminance
• Technically,
– Luminance = incoming light
– Brightness = perceived incoming light
• Many color models make the luminance or brightness an
explicit component of the color model though by varying
names:
–
–
–
–
–
–
Luminance
Intensity
Brightness
Lightness
Luma
Value
Chromaticity
• Requires two parameters
– Usually:
• Hue
• Saturation
The dominant wavelength
How pure that color is
(ratio of color to white)
– Some models use different chromaticity parameters
• Examples:
– Red vs. Blue?
– Red vs. Pink?
– Dark red vs. Bright red?
Tints and Shades
• Lightening or darkening colors:
– Tint: adding more white to a color
– Shade: adding more black to a color
• Tints and shades are not inverses!
– Why not?
The CIE RGB Model
• Standardized “red”, “green”, and “blue” to try to
match perception: 438.1, 546.1, and 700 mn
R = Km
G = Km
B = Km
∫ L ( λ ) r ( λ ) dλ
∫ L(λ) g(λ) dλ
∫ L(λ) b(λ) dλ
Notice negative weight
€ of red component for
wavelengths near 500 nm—can’t physically mix
The CIE XYZ Model (1931)
• Replaces R, G, B with “imaginary” primaries X, Y, Z
– Don’t correspond to single wavelengths
– Weights are always positive
– More saturated than
monochromatic light
– Can model all physically
realizable colors
X = Km
Y = Km
Z = Km
€
∫ L ( λ ) x ( λ ) dλ
∫ L ( λ ) y ( λ ) dλ
∫ L(λ) z (λ) dλ
Normalizing CIE XYZ Model
• Normalize X, Y, Z so that we can talk
about chromaticity:
X
X +Y + Z
Y
y=
X +Y + Z
Z
z=
X +Y + Z
x=
€
Notice that x + y + z = 1 forms a plane in color space
— can just use x and y (because z = 1 - x - y)
CIE Chromaticity Diagram
Gamut
Color Gamuts
• A color gamut is the space of colors spanned by a set of
primaries
• No three physical primaries can span the entire space of
physically realizable colors
– Use “negative” weights for some colors, or
– Use “imaginary” primaries that lie outside the color space
– Both are physically impossible
• Better with more primaries, but still not able to span the
entire color space
– Some color printers use more than just CMYK
Device Primaries and Gamuts
• Different devices have different primaries and
different gamuts
– Scanners
– Monitors
– Printers
• Two major problems working with color:
– Calibration: “talking the same color language”
– Out-of-gamut limitations
NTSC YIQ Model
• The National Television Standards Committee (NTSC)
standard uses a color model called YIQ:
– Y luminance
– I and Q chromaticity
Y  0.299 0.587 0.114R 
  
 
I  = 0.596 −0.275 −0.321G
Q 0.212 −0532 0.311B 
• Television Signals
– Color TV broadcasts are YIQ
€– “Black-and-white” (grey) TV uses Y only
HSI Color Model
• One axis is intensity (luminance)
• The plane perpendicular to this axis
represents chromaticity
– Angle = hue
– Distance from center = saturation
• Can map this plane using a triangle,
hexagon, or circle
HSI Color Model
HSI Color Model
HSI Color Model
Other Color Models
• Hue-Saturation-Value (HSV)
•
•
•
•
– Like HSI but with only one cone
Hue-Lightness-Saturation (HLS)
Hue-Value-Chroma (HVC)
CIE LUV
CIE La*b*
– Attempts to be perceptually linear
Pseudocolor (Indexed Color)
• For some applications, you may want to store fewer than
24 bits per pixel
– Older/less-expensive video buffers
– Compression (used in GIF compression)
• Use a lookup table to map values in the image buffer to
24-bit color values
– Example: 8-bit buffer, 256-color lookup table
– The set of displayable colors in the lookup table is often called
the palette
• Selection of palette from an image is known as color
quantization
Pseudocolor (Indexed Color)
• Can use similar
method for
mapping grey
level image to
color
Color and Visualization
• Color can be used in information visualization to
provide an additional dimension
Color and Visualization
• Specific color meanings: keep it limited
• Color scales are hard to interpret
– Heat scale is one that of the few that partially
works
• Limits of color reproduction can change
the message
Color Image Processing
• General approaches
– Process RGB color planes separately
• Simple
• Not always meaningful thinking in RGB
– Convert to HSI or other color space with explicit
intensity component
• Process the intensity component as you normally would
• May process the saturation
• Generally want to leave the hue alone in this model
– Design transformation in intensity/chromaticity space,
but implement through direct RGB manipulation
Processing RGB Planes
Processing Other Components
Color Transformations
Color Transformations
• Can change channels
together to transform
the intensities
Color Balancing
• Or you can change
channels separately
– Can emphasize or
de-emphasize one
or more channels
Color Histogram Equalization
• Convert to HSI or similar space
• Process
– Histogram equalize the intensity plane
– Increase the saturation
– Leave the hue plane unchanged
• Convert back to RGB for display
Color Histogram Equalization
Summary
• Pick a color space that makes sense for your application
• If the application is interactive, consider using HSI or
other color space with explicit intensity/chromaticity
• If you want to process intensities but keep hues the same,
work (or at least think) in that kind of color space
• Keep physical limitations in mind
– Primaries, different color gamuts, etc.
• There’s a lot more to color than RGB triplets!