[Data Visualization] Colors (Perception)

Why Color?

• Why is color in vis important?
  • Supports preattentive processing
  • High accuracy for visualizing ordinal or nominal data
• Limitations
  • Relative perception (color constancy, background is important!)
  • Color deficiency: color blindness (색맹) and color weakness (색약)

Color

• You can name each color easily, but there are complex things behind color perception
• Let’s start from physicists’ view.

Light as Electromagnetic Wave

• Light of a single frequency is perceived as monochromatic light (단색광).

• Light in the real world is a mixture of lights with different frequencies!
  • Power spectrum of light

Color is not Wavelength

• There are colors that are not monochromatic.
  • Unsaturated colors such as magenta, gray, or white.

Human Eyes

• Retina (망막) receives light and converts the light into neural signals.
  • The signals are transmitted to the brain for visual recognition through the optic nerve.
• How to convert light to signals?
• Rod cells and cone cells!

Color Vision

• Two different kinds of receptors on the retina
• Rods (간상세포)
  • Active at low light settings
  • Low-resolution black and white information
  • 100 million rod receptors
  • 杆= 몽둥이, 막대 = rod
• Cones ( 원추세포) for Color
  • Active at normal lighting conditions
  • Three types of cones (sensitive at a different wavelength)
  • 6 million cone receptors
  • Dense in the center of vision (fovea, 중심와)

• Cones are most dense at the center of vision or the fovea.
• So, color perception is most accurate at the fovea.

Three Types of Cone Cells

• There are three types of cone cells:
• Long: most active at 580 nm (red, 63%)
• Middle: most active at 540 nm (green, 31%)
• **Short: most active at 450 nm (blue, 6%)

• S is much less sensitive!
  • We are not good at perceiving blue!
  • Do not show detailed information (e.g., text) in pure blue on a black background.

Trichromacy

• Trichromacy (삼색형색각): the spectrum of light is reduced as three values (tristimulus values , responses from S, M, and L)
  • i.e., retina encoding

Effects of Retina Encoding

• Because the brain only processes the three dimensional tristimulus values of light, any spectra that create the same tristimulus response are indistinguishable.
  • i.e., metamerism, 조건등색

Color matching Experiments

• Are tristimulus values consistent over humans?
• In 1920s and 1930s, researchers conducted color matching experiments where an observer adjusts three primary lights to match each sample color.
• Who is an observer?
• There is no “representative” observer for humans.
• We will test a group of people, check if there is a consensus, and if there is, we will use the average of results.
  • i.e., standard observer

• How to generate color to match and three primaries?
• We use single-wavelength colors.
• A prism and a slit to select a narrow band of wavelengths as desired.

• Three primaries:
• red, green, and blue (700 nm, 546 nm, and 435 nm)
• Target color: random color other than the three primaries.

• The result of a single trial is the three tristimulus values (or weights) that an observer used to match the colors.
• We call such three values ҧ 𝑟 𝜆 , ҧ 𝑔 𝜆 )), and ത 𝑏 𝜆

Experimental Results

Why Negative Values?

• Some colors cannot be matched by adjusting the three primaries.
• This doesn’t happen in the color space (or gamut ) of your
  • Your monitor will show a color by mixing three “monitor” primaries.
  • Therefore, all colors that your monitor shows can be factorized into the sum of three primaries with positive weights.
• In the experiment, we test all single wavelength lights, so it can happen!
  • We cannot subtract a primary with a negative weight below 0.
  • All we can do is just completely turning off a primary.

• “Negative primaries” can be simulated by adding a primary to the test light!

rg Chromaticity Diagram

• Plot ҧ 𝑟 𝜆 and ҧ 𝑔 𝜆

CIEXYZ Color Space

• The international standard for color specification (1931, by CIE or International Commission on Illumination)
  • Motivation: we want positive values for all human visible colors.
  • This wasn’t possible with RGB primaries. So, we define new virtual primaries, X , Y , and Z
• We will linearly transform ҧ 𝑟 𝜆 , ҧ 𝑔 𝜆 )), and ത 𝑏 𝜆

• Finally, we get the xy chromaticity diagram.

CIE Chromaticity Diagram

• CIE chromaticity diagram encompasses all the perceivable colors in 2D space (x, y).
  • Where is pink light?
  • Where is black light?

Gamut

• The gamut or color gamut is a set of colors that can be reproduced by mixing the given primaries.
  • sRGB is the RGB space that you use (sadly, it is quite small).

RGB Color Space

• There are many RGB (red-green-blue) color spaces depending on how you choose the three primaries.\
• However, it is hard to imagine the actual color from the color code!
• sRGB(100, 150, 200)?

HSL Color Space

• HSL(hue saturation lightness) color space
  • More intuitive
  • Used by artists and designers
• Hue: what color (red, blue, green,…)
• Saturation: the amount of white mixed with the pure color
  • Pink = Red + Some amount of White
  • How colorful
• Lightness: the amount of black mixed with a color
  • How bright

• The HSL color space is more “interpretable” but it is pseudoperceptual
• It does not truly reflect how we perceive color.
• Especially, the lightness L is widely different from how we perceive luminance.

The Lab* Color Space

• The L*a*b* color space calibrates the limitation of HSL.
  • CIELab
  • Another CIE standard
• L*: perceptual lightness
  • L* = 0 for black L* = 100 for white
• a*: green red
• b*: blue yellow

• The L*a*b* color space approximates the perceptual lightness.
  • Blue: RGB(0, 0, 255) –> Lab(32 , 79, 107)
  • Yellow: RGB(255, 255, 0) –> Lab(97 , 21, 94)
  • Blue looks darker than yellow since its L* is smaller (0 = black).

• a* and b* in Lab* represent the green red and blue yellow components.
• Why these two axes were chosen?
  • Why not green yellow and red blue?
• It is based on the opponent color model of human vision

Opponent Process Theory

• Presented by German psychologist Ewald Hering late in 19th century.
• Supported by a variety of experimental evidence
• Colors are arranged perceptually as opponent pairs along three axes.
  • black-white (L*)
  • Green-red (a*)
  • yellow-blue (b*)

Spatial Sensitivity

• The red-green and yellow-blue chromatic channels are capable of carrying only about one third the amount of detail arried by the black white channel.

• The theory explains why there are “yellowish green” or “greenish blue” but no “reddish green” nor “yellowish blue”.

• The theory also explains why there are primary color terms consistent across different cultures and languages (Berlin and Kay, 1969).
• The first six terms define the primary axes of the opponent color model!

Color Deficiency

• About 10% of male and about 1% of female have some form of color deficiency.
• Red-green color deficiency (적록색맹)
  • Protanopia(lack of L, 제1적록색맹) and Deuteranopia (lack of M, 제2적록색맹)

• There are a lot of tools where you can check how your vis looks to people with color deficiency.

Summary : Colors (Perception)

• Rods for black and white, Cones for color
• Three types of cones: S (blue), M (green), L (red)
• Metamerism
• Color-matching experiment with “a standard observer” and three primaries
  • Negative weights (for light whose wavelength is ~500 nm)
• CIEXYZ calibrates these negative weights and assigns positive weights XYZ to all human visible colors.
• CIELab models a perceptually uniform color space.
  • L: black-white, a: green-red, b: blue-yellow
• Opponent Process Theory and color deficiency