Post

8. Map Color and Other Channels

Posted Updated
By Mingyu An
8 min read
8. Map Color and Other Channels

Design Choice for Mapping → Colors

Rods and Cones

  • Rods
    • 막대세포
    • active at low light levels
    • one wavelength sensitivity function
    • 100 million rod receptors
  • Cones
    • 원뿔세포
    • 3 cone response signals
    • three types
    • 6 million
    • focused in the center of vision (fovea)

Terms

Luminance (조도)

  • Physically measured amount of light (lux로 측정되는 그것)

Brightness (밝기)

  • Perceived amount of light
  • non-linear (adaptation, contrast)

Lightness (명도)

  • Perceived refelctance of a surface
  • depends on a reference light
  • white surface is light, black surface is dark

Hue

  • color

Saturation

  • Perceived intensity of specific color
  • vividness

We change meters

eye sensitive over 9 orders of magnitude

5 orders of magnitude → room ~ sunlight

  • Receptor bleach and become less sensitive

→ We’re not light meters. 우리는 Meter를 바꾼다.

  • 신경체계는 신호의 차이를 계산한다.
  • 정확한 숫자 값을 계산하지 않는다.

Color Vision

Rod: black-white. Low light settings

Cone: color, normal light setting

  • Three opponent channel
    • Red-green
    • Blue-yellow
    • Black-white
  • Luminance: high resolution
  • Chromaticity: low resolution

Color Space

Mathematical ways to describe color

  1. RGB: computationally easy. poor mechanics
  2. HSL: Hue, Saturation, Lightness
    • Pseudo-perceptual
  3. HSV: Hue Saturation Value(for gray scale)
  4. CIEXYZ → 인간의 원뿔세포를 모델링하여, 인간이 인식 가능한 색상을 표현하기 위한 것이 있음.

Three Independent Channel

Retina contains three type of color receptors with different absorption spectra

Input * Con → Three Values

자극 그래프와 각 response 곡선을 곱하고, 적분하여 전체 값을 구한다.

Ambiguity in trichromacy

→ input stimuli가 달라도 최종 인식된 값을 같게 만들 수 있다.

matamerism.

Model of Standard Observer

Color matching experiment

이 자극이면 평균적으로 이런 r,g,b를 갖더라.

→ Color matching function 정의.

Primary Colors of Light

  • Red, green and blue (원뿔세포의 종류에 따라 r, g, b에 가까운 색에 반응하기 때문)
    • L cones: 65%
    • M cones: 33%
    • S cones: 2%

→ Color matching experiment에서 R을 음수로 해야하는 경우가 있음. (반대쪽에 R쏴주는 식으로 구현)

CIERGB Color System

  • Based on experimental results
  • R, G, B color를 사용하면 negative primary color를 사용하는 구간이 존재한다.

→ Normalize

경계는 단색광. (보라색 제외)

사실 이 map은 perceptually uniform하지 않다.

→ equally size step appear equal to our visual systems. (CIElab, CIEluv)

Color Space

HSL: popular color space.

  • pseudo-perceptual
  • L is different from true luminance

→ L_a_n*

좀 더 perceptual한 것을 반영.

Color Theory

Luminance: physical light

  • Brightness: Perceptual light

Saturation: Perceived intensity of specific color

Hue: Wavelength

Luminance

  • As magnitude channel?
    • Suitable for ordered data types
    • But not accurate → quantitative에 부적합함.
      • 많아야 5개임.
    • resolving detail을 위해 밝기 대비.
      • luminance가 비슷하면 글도 안읽힘. 3:1은 기본이고, 작은 텍스트라면 10:1

→ Luminance is different from brightness

  • Brightness(밝기)는 인식된 것임. context와 환경에 강한 영향.

Saturation

Magnitude of saturation

  • Low accuracy. 3단계 정도.

크기 채널과 관련 있음

  • 사이즈가 찾을 때 saturation 알기 어려움
  • 밝고, 선명한 컬러는 작은 영역에
  • 흐린 파스텔 톤은 넓은 영역에

Hue

as an identity channel

  • categorical data, grouping에 효과적.
    • Spatial position 다음으로 효과적임 (nominal 한정)
    • 특정 순서 없음. (빨주노초파남보는 학습된 것임)
  • 크기 챈널과 역시 인터랙션함
    • 작으면 색깔 안보임
  • Continuous region에서 fine distinctions.
  • discontinuous에서는 lower accuracy. 약 6~7레벨

Cross-Cultural color naming

→ consistent across culture

white/black → red → ( green or yellow ) → blue → brown → ..

Color Categories

Task: Name the colors!

Only 8 hues named out of 210 colors. Small # of labels.

Color Coding

Large areas: low sat. enough

Small areas: high sat is needed.

→ Break isoluminance with borders!

Must have luminance contrast with background for details.

Transparency

Interacts with other three channels

  • luminance, saturation, (hue). 투명도 바꾸면 이것도 사실상 바뀌는 것임.
  • Superimposed layers에 자주 사용.
    • glassmorphism도 이와 유사한 듯.

Colormaps

speicifies a mapping between color and value

  • Categorical
  • Ordered (sequential, diverging)
  • Continuous
  • Segmented

Color Brewer

  • Sequential scheme
    • dark for high
  • Diverging scheme
    • mid-range critical values and extremes at both ends of the data range
  • Qualitative scheme
    • difference between class

Categorical Colormap

  • Encode categories at colors
  • 6 ~12 color max.

Difficulties in categorical colormaps

→ color should be close in luminance (major difference in salience, ensure against background)

→ color should be different in luminance (distinguished even in black and white, color blind)

Resolve

  • Distinguish bars in large region next to each other. vs distinguising non contiguous small regions

→ not enough colors?

  1. reduce # of bins
  2. Use additional visual channels

Ordered Colormap

Rainbow colormap is poor. 😟

  • Diverging: Two hues at end points, with neutral at midpoint
  • Sequential: Grayscale / Control saturation and brightness og single hue

How many hues to use in ordered colormaps?

→ Depends on what level should be emphasized (high, mid, low…)

  1. Many hues → Mid-level neighborhood structures are emphasized.
    • Low-frequency are highlighted
      • segmentation
    • Easily discuss subranges
  2. Two hues (diverging) → High-level structures are emphasized
    • Continuity
    • Diverging color map

Again, Rainbow colormap is poor!

easily namable color…but

  1. Hues don’t have implicit perceptual order
  2. Scale is not perceptually linear.
  3. Fine detail can not be perceived with only hues

→ 사실 노란색 부분이 의미 없을 수 있다.

→ 데이터 semantic 구조와, perception 구조가 다를 수 있다.

따라서, 단조 증가 luminance colormap을 사용하고, multiple hue를 semantic에 맞게 사용하자.

  • popular interpolation은 perceptually nonlinear하다
    • 물론 이걸 linear하게 바꾼 연구는 있지만, 칙칙해서 안쓴다.

Then how to use?

  • Segmented rainbow colormaps are good categorical data.
  • Deliberately bin the data explicitly
  • or let eyes create bins

Color Deficiency

  • Deuteranope (green defict)
  • Protanope (red-green)
  • Tritanope (blue-yellow)

Why color blindness?

→ red green, yellow blue, black white 고장나기 쉬움

  • Most colorblind는 적록색맹이다. red-green color scheme을 쓰지 말자.

Size Channel

  • effective for ordered data
  • Affect most channels
    • saturation, lhue
    • shape
    • orientation
  • 길이는 정확한, 넓이 부피는 부정확함
  • 더 큰 채널은 subsumes함

Angle Channel

  • Orientation of a mark
  • Sequential, diverging, cyclic
  • Accuracy is not uniform!
    • vertical, horizontal, diagonal은 정확함
    • 하지만, 다른건 아주 형편이 없음

이런 orientation resolution은 45도일 때 최대임. 즉, 90도에 가까이 하면 이게 올라가는지 내려가는지 알기 어려움. aspect ratio가 중요하다 이말이다.

Curvature, Shape Channel

  • Curvature: 아주 부정확함. 쓰지 마라.
  • Shape:
    • point and line 위에서 identity channel로 작동 가능
    • size와 shape에 strong interaction

Motion Channel

  • Direction , velocity, frequency
  • Salient, separable from other states
    • Draws too much attention.
      • 그래서 bi-state로 쓰는게 좋음. (움직인다, 안움직인다)
  • Used for highlighting
    • blinking (의료 도메인에서 변화 파악)
    • transitory vs ongoing

Texture and Stippling Channel

Texture

  • Very small-scale pattenrs
  • Categorical attributes (order도 가능.. density 조절하면 됨)

Stippling

  • Fill in regions of drawing with small strokes
  • Used at older printing tech. (점묘화 같음)
Lecture Notes, Information Visualization and Visual Analytics
info-vis visualization color design channel
This post is licensed under CC BY 4.0 by the author.