---
theme: custom-theme
---

# Perception & Color

## CAPP 30239

---

## Today

- What matters most when creating a visualization?
- How does human **perception** factor into visualization design?
- Understanding **color**, and computational representations of it.

---

## What is the most important question when creating a visualization?

---

## What is the most important question when creating a visualization?

<ul>
<li><s>Where will the data come from?</s>
<li><s>What type of chart do I use?</s></li>
<li>Who is the audience?</li>
</ul>

---

## Audience First

- Who are you presenting to?
- How familiar are they with the data?
- What is their numerical & visualization literacy?
- Via what medium will they receive the information?
- What are you trying to do? (Persuade, Inform, Inspire?)

*Only now can we start thinking about data and presentation.*

---

## Perception

- **Selective** - We can only pay attention to so much.
- **Patterns** - Our brains are pattern-matching machines, audience will benefit from intentional patterns & be distracted by unintentional ones.
- **Limited working memory** - We hold a very limited set of information in our minds at once.

---

## What do you see?

<div class="container">
<div class="col">

![](viz-1.png)

</div><div class="col">

```python
alt.Chart(random_df).mark_point().encode(
    alt.X("a"),
    alt.Y("c"),
    alt.Color("b"),
    alt.Size("c"),
    alt.Shape("a:N"),
    alt.Fill("b"),
    alt.Opacity("b"),
) 
```
</div>
</div>

---

## What do you see?

<div class="container">
<div class="col">

![](viz-2.png)

```
alt.Chart(random_df).mark_line().encode(
    x="a",
    y="c",
)`
```

</div>
</div>

---

## Effectiveness Revisited

![width:800px](effectiveness.png)

---

<div class="container">

<div class="col">

**Altair Channels**

- Position (`X, Y`)
- Angle (`Angle`)
- Area (`Radius`, `Size`)
- Hue, Saturation (`Color`)
- Texture (`Opacity`, `Fill`)
- Shape (mark type, `Shape`)
</div>

<div class="col">

**What about?**
- Length
- Slope
- Volume
- Density
- Connection
- Containment
</div>
</div>

---

**Derived Properties**


- Length/Area - size of bars (`X`, `Y`)
- Slope & Density - affected by scale
- Connection - ex. layering of lines w/ points
- Containment - achieved with layering

What about *volume*?

---

## Stevens' Power Law

Stevens (1975): Human response to sensory stimulus is characterized by a power law with different exponents with different stimuli. 

perception = (magnitude of sensation)<sup>a</sup>

Smaller <sup>a</sup> exponent: harder to perceive changes.

Stevens measured values of a by exposing people to varied stimulus and asking them to compare magnitudes.

---


<div class="container"><div class="col">

![](stevens.png)

</div><div class="col">


| Continuum | Exponent |
|-|-|
| Color **Brightness**| 0.33-0.5 |
| Smell| 0.6 |
| Loudness | 0.67 |
| **Depth Perception** | 0.67 | 
| Area | 0.7 |
| 2D Planar Position | 1.0 |
| Warmth | 1.3-1.6 |
| Color **Saturation** | 1.7 |
| Electric Shock | 3.5 |

</div></div>

---

## 3D Graphs

![](stunning-3d-chart.jpg)

---

![](datavizproject.png)

---

![](3d-scatter.png)

---

## Instead of 3D Graphs

- Find other channels: hue & size are good candidates. (bubble chart)
- Or make multiple 2D graphs with XY/YZ/XZ pairs.

![bg left](scatter-matrix.png)

---

## What is Color?

Wavelengths of light are perceived as particular colors:
![](linear_visible_spectrum.svg.png)

What's missing?

<!-- credit: https://en.wikipedia.org/wiki/File:Linear_visible_spectrum.svg -->

---

## Color & the Eye

### Rods

- spread throughout retina
- more sensitive in low light conditions
- brightness ("lightness")

### Cones

- 3 types with peak sensitivity at different frequencies
- concentrated in center of eye
- less sensitive in low light conditions
- hue & saturation

![bg right](cone-fundamentals.png)

<!-- source https://commons.wikimedia.org/wiki/File:Cone-fundamentals-with-srgb-spectrum.svg -->

---

## Spectrum vs. What We See

What we actually see is always a blend of multiple peaks.

This is impacted by ambient light conditions, as well as quirks of our visual processing.

![height:400px](the-dress.jpg)

In actuality, multiple combinations of light can give same color (**metamers**).

---

## Chromatic Adaptation

![](demo-bw-illusion-andrew-steele.gif)

Source: Andrew Steele <https://www.youtube.com/channel/UC-XYsDNh4-886rMNLnnwR_w>

---

## Color Naming

Color naming is highly subjective, and research has shown that the ability to name a color correlates highly with the ability to distinguish it.

![bg right](xkcd-color-map.png)

Be particularly careful with blue/green boundaries, as there are significant cultural differences.

Source: https://blog.xkcd.com/2010/05/03/color-survey-results/


---

## Cultural Considerations

![bg right width:600px](hok-uk.svg)

- American audiences associated <span color="red">red</span> & <span color="blue">blue</span> with political parties on any map in a political context.
- Also international meaning of <span color="red">red</span> & <span color="blue">blue</span> is flipped: red is left, blue is right.
- Most other colors have contradictory meanings depending on culture. For example, yellow might be chosen to denote success (parts of Africa) or be associated with death (Middle East).

<!--Source: https://www.color-meanings.com/color-symbolism-different-cultures/-->

<!-- image from wikipedia: UK House of Commons -->

---

## Color Vision Deficiency

More accurate name for what is commonly known as colorblindness.

- Red-Green CVD - most common
    - four types: Dueteranomaly and Protanomaly (mild) to Protanopia and Dueteranopia (complete)
- Tritanomaly/Tritanopia: blue/green and yellow/red confusion.
- rarest, complete lack of color vision, usually corresponds to other vision issues as well

![bg right width:600px](colorblind.jpg)

---

## Color on a Page

![bg right](cmyk.svg)

Ink absorbs light, so we work with subtractive blending.  Our base colors are cyan, magenta, and yellow.  To save on ink costs, we throw in black/contrast as well.

We call this CMYK color.

---

## Color on a Screen

Screens emit light, which means we use **additive blending** of red, green, and blue light. Every pixel of a screen can emit these three colors in different intensities.



![bg right](additive.png)

---

## Color Spaces

Ways of describing a color mathematically, usually have 3 components to match our perception of color:

- RGB (early photography)
- CIE XYZ (1931)
- HSB/HSV/HSL (1970s)

![bg right width:700px](rgb-khan.jpg)

<!-- https://commons.wikimedia.org/wiki/File:Rgb-compose-Alim_Khan.jpg -->

---

A common way to refer to colors is by their intensity in each of these three channels.

<span style="color: rgb(0% 100% 0%)">this is 0% red, 100% green, 0% blue intensity (#00ff00)</span>
<span style="color: rgb(20% 60% 20%)">this is 20% red, 60% green, 20% blue intensity: (#143c14)</span>
<span style="color: #ff00ff">this is 100% red, 0% green, 100% blue intensity: #ff00ff</span>

This is sometimes expressed in hexadecimal:

![height:120px](hexrgb.png)

![bg right](rgb-pixels.jpg)

---

### RGB space as a cube

![cube](LinearRGBCube.png)

---

### RGB as pair plots

![pair plots](RGBPairPlots.png)

Remember this trick for your own 3-dimensional data!

---

![height:500px](TriangleSliceRGB.png)

A slice through the middle of the cube gives colors of comparable brightness.  (You may have seen such a triangle in color pickers.)

---

## HSL

![height:500px](HSL.png)

An alternative color space that's very useful for visualization is HSL color space.

Hue, Saturation, Lightness | <https://hslpicker.com/>

---

## Aside: What about "alpha"?

You will often see a fourth channel: RGB**A**, HSL**A**.

This is known as alpha transparency (translucency).

This has to do with how the program in question *blends* the colors. The final pixel values on the screen will still be converted to RGB components.

- Use sparingly.
- Variations are very subtle, and background dependent.

<!-- image source: https://upload.wikimedia.org/wikipedia/commons/3/34/RGB_pixels.jpg -->

---

## CIE (RGB / XYZ / CIELAB)

Based on human perception experiments where people would adjust dials to recreate colors out of red, green, and blue light.

First from 1920s, revised in 1970s.

*Commission internationale de l'éclairage*  (Illumination)

![bg right](cie-xyz.png)

---

## Screen Gamut

Screens can't show the entire range of visible colors accurately, they define a "gamut".  Since ~1996 most devices aim at a standard gamut to ensure similar representations of color, but even high end devices are not perfectly aligned.

![bg right](gamut2.png)

Projectors (like the one you're likely viewing this on) usually have skewed gamut. 

Moral of the story: **Consider your medium!**

---

## What does all this mean for visualization?

Color choices should be made with respect to:

  - medium (screen vs. print, type of screen)
  - audience (culture, vision differences, expectations)
  - differentiability

---

## Role of Color

- **Identify** - Different color per category/actor.
  - opt for distinct hues
- **Group** - Group like entities using same/like colors.
  - often with similar hues
- **Layer** - Overlay different information while keeping contrast.
  - saturation differences very important to not overwhelm eye
- **Highlight** - Call out important/relevant information.
  - brightness and hue differences important

--- 

## Color Channels & Data Types

### Lightness is perceived as ordered

Good for **Ordinal** variables

![height:50px](ordinal.png)

**Quantitative** (Continuous) variables harder to discern

![height:50px](qual.png)

### Hue typically unordered

**Nominal** variables.

![height:50px](nominal.png)

---

### Types of Palettes

- Qualitative - Nominal data
- Sequential - Quantitative data
- Diverging - Data with a meaningful zero-point (increase/decrease, more/less)

![](palette-types.jpg)

<!-- source: Peter Aldhous, NICAR 2016 -->

---

### Hue Separation

Pick distinct hues for unrelated variables.

Grouped schemes can be used where there are relationships among the categories.

![bg left width:600px](vega-schemes.png)

<https://vega.github.io/vega/docs/schemes/>

---

## Color Tips

- Aim for no more than ~6 colors that need to be distinguished.
- Colors should be distinct & differentiable by name.
- Be mindful of cultural considerations & symbolism.
- Ensure color schemes chosen appropriately for types of data.
- "Get it right in black & white"


---

## "Get it right in black & white"

A common mantra among visual designers.

Ensure that your hues have different brightness levels.

Ensure that you aren't using hue alone for your image.

![bg left width:600px](vega-schemes-bw.jpg)

---

## Text Legibility

An important issue when using colored text and/or backgrounds is **legibility**.

Web Content Accessibility Guidelines require a 4.5 color contrast (3:1 for large text).

Minimize *saturation* in backgrounds, pick a font color with opposing *lightness*.

<https://webaim.org/resources/contrastchecker/>

![bg left width:600px](contrast.png)


---

## Tools

- Vega Schemes: https://vega.github.io/vega/docs/schemes/
- Contrast/theme exploration: https://schubert-da.github.io/dataviz-palette-tool/
- Theme exploration for cartography: <https://colorbrewer2.org/>
- Color-theory based theme creator: https://meodai.github.io/poline/
- Theme creator w/ theme sharing: https://coolors.co
- HSL/RGB picker: <https://hslpicker.com/>
- Contast checker: <https://webaim.org/resources/contrastchecker/>

### Color-Blindness

- MacOS/iOS app: https://michelf.ca/projects/sim-daltonism/
- Browser extensions (search "colorblindness" in your browser of choice)

---

## Acknowledgements & References

Thanks to Alex Hale, Andrew McNutt, and Jessica Hullman for sharing their materials.

Color space images are from <https://jamie-wong.com/post/color/>, which is an incredible resource if you'd like to go deeper into both the biology and math of color.

- https://www.math.csi.cuny.edu/~mvj/GC-DataViz-S23/lectures/L6.html
- https://en.wikipedia.org/wiki/Stevens%27s_power_law
- https://colorusage.arc.nasa.gov
- https://vega.github.io/vega/docs/schemes/