Why talk about visual perception & Encoding?
Parametrize a preset chart
e.g. Which color to use?
Adjust a base chart
e.g. to counter a perceptual bias
Built own chart types
To tailor it to a particular dataset
Need to detect when someone tries to deceive you with a chart
We are able to know how a chart was built, and try to detect if it is proper to
build in this way, it is very helpful while someone trying to deceive us with a
chart. (Visual Literacy)
1. The human eye, visual perception & cognition, illusions
& biases
Some implications of color:
If we want to highlight something important, we should use red, cuz our eyes
will be drawing automatically to red and red is easier & faster to perceive. We
can also make the effect stronger if we reduce all the other colors (set others
to grey)
Only 2 percent of the cones are actually for blue perception, so blue is rather
hard to perceive.
Changes between brightness are even easier to perceive than changes in color
hue. We can use luminance contrast boundaries to offset different colors, to
make it stand out from the background, and easier to perceive.
What is color? ⾊调，饱和度，亮度
Color = Composite of Hue, Saturation and Brightness
Not formal: Tint&ShadeBrightness; Tone Saturation
Luminance: “Perceived” Brightness, not just how “light” they are.
Pure color(Pure hue/Fully saturated colors/Do not consider about tone or
saturation) already have different luminance. Luminance is an inherent
property of pure color.
About luminance and RGB: Green has the most weight while Blue has the least
Luminance = visually perceived brightness
sometimes called color temperature
Contrast: Difference in luminance
着⾊
Pre-attentive visual cues
Pre-attentive means that you can without really having to read the visualization,
you can immediately just look at it and see immediately what it is about.
Position, color, and size are pre-attentive visual cues. It doesn't take u any
effort to read the chart and find the outlier. Position always the most effective.
Other cues: Length, Density, Luminance, Shape, Orientation, etc.
Luminance> Hue, Hue>Form/Shape
格式塔原则
Gestalt Principles of Grouping
A principle for grouping
The way of how to show items belongs to each other
外壳
接近
闭合
Proximity: Items in the same group are closer (Position-wise)
Enclosure: circle the items to show that they belong to each other
Cognitive Distortion and Illusion
Certain algorithms to correct certain illusions, but you need to make sure your user will
not correct the illusion by themselves.
Bias & Anchoring effect on visualization
锚定效应 Anchoring effect is a cognitive bias where an individual depends too heavily on
an initial piece of information offered to make subsequent judgments during
decision making.
Be aware of this bias when you doing the visualization, give them new visual
experience and try to get the users out of their “pre minds” and let the decouple the
ones they saw before.
2. Marks and Channels
Marks: Visual objects (Lines, Points, Areas, etc.)
Channels: Properties of marks. (shape, size, position, color, etc.)
Utilize the properties/Map data to visual channels according to data types.
Data types
Quantitative Data: Continuous data + Discrete Data
Qualitative Data: Ordinal + Categorical Data
Mapping Criterion
Expressiveness: use the “right” channel to encode data, let them make sense.
A visualization is expressive if it encodes all the data relations intended and
no other data relations.
Examples: Shape Categorical data; Length Continuous data; Color
brightness Ordinal Data…
Effectiveness: Which channel to faster, more easily to perceive for the
information in the visualization?
A visualization is effective if it best exploits the capabilities of the output
medium and the human visual system. (Make the visualization easy & fast to
perceive)
Appropriateness :A visualization is appropriate if it balances the tradeoff
between efforts required for creating it and the benefits yielded by it.
Cost/Gain
For Effectiveness
Note: Nominal=categorical
Position always the easiest to perceive, so usually for most important information.
Shape always bad for sortable data.
3. Marks: Dimension
The perils of 3D: 危害
Occlusion 遮挡
Perspective Distortion
Non-anchored points (e.g. Can not read tick marks for a points without interation)
Navigation/Orientation (Lost in Space/Can not place the users)
Selection
Technical Issues (e.g. Font Renderings. Subpixel rendering only for 2D)
If you want to go beyond 2D, you can use 2 and a half D.
E.g. 2D backgrounds, 3D objects.
[Link]: Color
Equidistant
Equidistant means that the perceive change in color should be equivalent to the
real change of the amount of the data. Luminance increasing evenly.
What color scale we want for quantitative data?
Increasing evenly, continuous luminance, equidistant.
Why rainbow color scale is heavily criticized?
(rainbow color scale: based on rainbow, order according wavelength.)
Not equidistant. Two peak in mid of luminance. We want luminance
increasing evenly
Alternative: Cubehelix color map
Note:
On Viz criticize, notice the color if they are:
Color blind-friendly, evenly increasing luminance/equidistant, accidentally
highlight?
Diverging color scale 中性的
When you have data that has a neutral point in the middle and then goes either
negative or positive
Multi-Hue and Single Hue
Multi-hue color maps are used when u want viewers to read off the a certain value
from the color.
Some reference for multi-hue:
Paul Tol’s (Quantitative and Qualitative) color
5. Labeling
Favor direct labeling over legends
Label Types:
difficult to
read
偏⼼
An algorithm for putting label : particle-based labeling.
Stage 1: Try putting a plain label for a point. 4-Position model8-position
modelThe slider model
Stage 2: If all positions are taken and we cannot put a plain label, try excentric
label. Move the label along a spiral path until a suitable position is found.
How to detect collision: Particles.
Selection of fonts:
No serifs. 衬线
No text figures
Rendering
Illustrative visualization