Color Physics
Simplified rendering model: illumination * reflectance (relative energy, 0 to 1) = color signal.
For project: use consistent light source to prevent color shifting
Human vision system: optimized for spectrum from the sun (greatest intensity at ~500 nm).
Reflectance spectra: reflectance (0 to 1) as a function of wavelength. Eyes (and cameras) simplify this signal by reducing it to intensity detected via three types of sensors and hence, objects with different spectral albedoes may be perceived as having the same color (metamerism).
Spectral colors can be described as a spectra with a single peak in some given wavelength range.
Color mixing:
- Additive color mixing: colors combine by adding intensity (e.g. displays)
- Subtracting color mixing: colors combine by multiplying intensity (e.g. paint)
Color Perception
The Human Eye
Specifications:
- Spectral resolution: 400 - 700 nm
- c.f. ears which have a far wider range of 10 octaves (20 Hz to 20 kHz)
- Dynamic range: approx.
, similar to ears - Spatial resolution: approx 1-3 cm at 20 m
- Radiometric resolution (differentiating between two colors after a time gap): around 16 - 32 shades of black and white, 100 colors
Physiology:
- Lens and fluid causes signal attenuation
- Rods (black and white)
- ~ 100 million rods
- Extremely sensitive: can be excited by a single photon
- ~170 degree field-of-view per eye
- Cones (color):
- ~ 6.5 million cones
- Much larger than rods
- A vast majority centered around the fovea
- Three types:
-
and spectra have large overlap -
cones much less sensitive than the other two types - Cameras usually require IR filters and are in a Bayer pattern (two green for each red or green)
- Dark-adapted (scotopic) vision more sensitive to shorter wavelengths (almost blind to red)
- Light-adapted (photopic) vision sensitive to all visible wavelengths, peaking at 555 nm
Brain and processing:
- Signals from left visual field of each eye go to right visual cortex and vice-versa
- ~800,000 nerve fibers but ~100 million receptors; one fiber carries signals from multiple rods/cones
Color Spaces
Although perceived color depends on illumination and surroundings, assume for now that the spectrum of light arriving at the eye fully determines the perceived color.
- CIE 1931 XYZ color space
- Subjects would vary intensity of RGB lights to match test color, adding color to the test color if necessary (leading to negative values for red)
- Perceptually uniform color space: any change of a given distance in the color space looks the same to the human eye regardless of the starting color?
- Not great for computer vision
- HSV cone:
- Angle = hue
- Radius = saturation
- Value = height
- At the bottom is the tip of the cone - black, which cannot have zero saturation
- Recommended color space: CIECAM02 (JCh)