1. CSc4730/6730 Scientific Visualization
Lecture 22
Cognitive Basis of
Information Visualization
Ying Zhu
Georgia State University

2. References
Jeremy M. Wolfe, How Might the Rules that Govern Visual Search Constrain the Design of Visual Displays? Society for Information Display Symposium Digest of Technical Papers, 36(1), pp
Colin Ware, “Information Visualization”, Morgan Kaufmann, 2nd edition, 2004.
Chapter 5 “Visual Attention and Information that Pops Out”

3. Outline
Visual search
Visual attention

4. Overview
One of the goals of data visualization is to effectively guide reader’s attention to important information
In visual search attention can be guided toward a target item among distracting items by a limited set of basic features.
This paper will summarize the rules of effective guidance

5. Basic visual search
If an observer is asked to find a target among a number of similar distracters that observer will behave as though she is processing those items at a rate of one every msec

6. Basic Visual Search
This assumes that the observer does not need to move her eyes to each item in turn because eye movements proceed at a much slower rate of 3-4 sec.
It also assumes that each item can be recognized rapidly once attention is directed to it.

7. Basic visual search
This is "face-in-the-crowd" search where one face is much like all others.
Why don't we have to search in this laborious fashion most of the time?
The core of the answer is that we can use simple information to guide our attention to more complex targets

8. Guided search
If basic feature information differentiates the target from, at least, some of the distracters, search can be speeded.

9. Guided search

10. Basic features
Only a limited number of stimulus properties can guide attention.
The exact list is a matter of debate but there are probably not fewer than a dozen guiding features and not more than two dozen

11. Basic features
The most efficient searches are those with a target item defined by a single basic feature
The target is different from distracters by a single feature

12. Basic features
Some features are fairly obvious candidates like color and orientation.
Others, like line termination are less obvious and many, seemingly reasonable, candidates like intersection type do not work

13. Basic features

14. Top-down vs. Bottom-up guidance
Basic features in a display can guide attention in either of two modes.
Top-down guidance
Bottom-up guidance

15. Top-down vs. Bottom-up guidance
Top-down guidance is based on the observer's attentional set (e.g. "look for red").
The reader knows what feature(s) to look for

16. Top-down vs. Bottom-up guidance
An item might attract attention without the observer needing to know anything about that item in advance.
This is bottom-up guidance and is based on the local differences between an item and its neighborhood

17. Top-down vs. Bottom-up guidance
Strong bottom-up guidance is sometimes referred to as "attentional capture".
However, you cannot assume that your flashing ad will grab eyeballs on a webpage.
Observers can usually learn to avoid capture by new stimuli if they form a top-down attentional set to do so.

18. Similarity
If you are trying to attract attention, it is important to keep in mind that bottom-up guidance is based on differences.
The larger the target-distracter differences, the more salient the target.

19. Similarity

20. Similarity
The larger the differences among distractors, the less salient the target will appear.

21. Similarity
This is why you want to limit the number of colors you use in data visualization
Rainbow color coding may look pretty but make visual search slow

22. Categorical processing
Top-down guidance has a very limited “vocabulary”.
Consider orientation. While it is easy to tell if a line is vertical, it is not easy to find a vertical line among lines tilted 20 deg to the left and right

23. Categorical processing

24. Categorical processing
It is easier to ask that attention be guided to items that are defined categorically.
In data visualization design, limit the number of variations in visual feature (color, size, orientation, etc.)
Use a small visual “vocabulary”
In orientation, the categories are steep and shallow, left and right.

25. Categorical processing
Search for the 10 deg tilted item is fairly easy since it is the only “steep” item

26. Categorical processing
Search for the 10 deg tilted item is harder because it is neither uniquely “steep” nor uniquely “left-tilted”.

27. Categorical processing
In size, the guiding vocabulary seems to be just “big” and “small”.
It is hard to find the medium-sized target in the following picture.

28. Categorical processing
This does not mean that multiple orientations, sizes, colors, etc. cannot be used in a display.
However, the choice of features is constrained when it is important that users be guided to one item without being distracted by others.

29. Crowding
Just as the deployment of attention is coarsely coded to a feature attribute like orientation, it is also coarsely coded in space,

30. Crowding Look at the center of Figure Five where it says "Look Here".
If you now attend to the pattern at the top of the figure, you will probably find yourself unable to count the lines.

31. Crowding
If you keep your eyes in place and shift attention to the letters at the bottom of the figure, you will find yourself unable to read the letters in the middle of the array.
As with the lines, you are unable to restrict your attention to a single letter when it is crowded by others.

32. Crowding
In data visualization, labels should be placed as close to the visualization items as possible

33. Combining features
We can guide attention based on the combination of multiple sources of information.
Conjunction search
Under most real world situations, knowing that a target is a moving, big, red, shiny object is likely to guide attention to only a few items even if the specifications of "moving", "red" etc. are quite crude.

34. Combining features
Under most circumstances, the crude descriptors will limit the set of possible targets to a small subset of all items.
The exceptions are the "needle in a haystack" searches where all items have much the same set of features.

35. Combining features
Guidance to conjunctions is also subject to some fairly severe limitations.
Most notably, we do not appear to be able to guide attention to the conjunction of two values drawn from the same feature space
Color-color conjunction
Shape-shape conjunction
Size-size conjunction

36. Combining features
On the left of figure, it is hard to find the target defined by a conjunction of two colors, red & blue

37. Combining features
Interestingly, it is easier if one color is the color of the whole item and the other color is the color of a part (The red item with the blue center in Figure)

38. Combining features
Be careful when you code multiple colors in one visual object
Colored stacked bar chart
Colored pie chart
Colored line chart, etc.

39. Objects
All else being equal, attention is guided to objects and that attention spreads throughout an object
(though the precise definition of objects is not clear)
However, it proves to be very difficult to guide attention to a specified part of an object.

40. Objects
It is hard to find green horizontal when it is embedded in objects that all contain green and vertical (left).
It is easier if there is only one green horizontal object (right).

41. Objects
If you encode information in a part of an object, it may be too subtle to be noticed

42. Visual attention
Attention is both a low-level and a high-level property of vision
Here we focus on low-level mechanisms that help us understand what is more readily available to attention

43. Visual attention
Experiments have shown that humans do not perceive much unless we have at least some expectation and need to see it
In most systems, brief, unexpected events will be missed

44. Useful field of view
Useful field of view (UFOV) is the size of the region from which we can rapidly take in information
UFOV varies greatly
With greater target density, the UFOV becomes smaller and the attention is more focused
With a low target density, a larger area can be attended

45. Tunnel vision and stress
In tunnel vision, the UFOV is narrowed so that only the most important information, normally the center of the field of view, is processed
When cognitive load goes up, the UFOV shrinks

46. Motion in attracting attention
Studies have shown that UFOV function can be far larger for detection of moving targets than for detection of static targets
Subjects can respond in less than 1 sec to moving targets 20 degrees from the line of sight
If static targets are used, performance falls off rapidly beyond about 4 degrees from fixation

47. Pre-attentive Processing
We can do certain things to symbols that make it much more likely to “pop out”
Pre-attentive processing determines what visual objects are offered up to our attention
An understanding of what is processed preattentively is probably the most important contribution that vision science can make to data visualization

48. Example

49. Example
The time taken to find the target is independent of the number of distracters

50. Why is this important?
In displaying information, it is often useful to be able to show things “at a glance”.
If you want people to be able to identify instantaneously some mark on a map as being of type A, it should be differentiated from all other marks in a pre-attentive way

51. Example

52. Pre-attentively processed features
Form
Line orientation
Line length
Line width
Line collinearity
Size
Curvature
Spatial grouping

53. Preattentively processed features
Blur
Added marks
Simple numerosity (e.g. less than 4)
Color
Hue
Intensity

54. Preattentively processed features
Motion
Flicker
Direction of motion
Spatial position
2D position
Stereoscopic depth
Convex/concave shape from shading

55. Preattentive symbols
Preattentive symbols become less distinct as the variety of distractors increases
Two factors are important in determing whether something stands out preattentively:
The degree of difference of the target from the nontargets
The degree of difference of the nontargets from each other

56. Which visual dimensions are preattentively stronger?
It’s hard to answer this question because it always depends on the strength of particular feature and the context

57. Coding with combination of features
What happens if we wish to search for a gray square, not just something that gray or something that is square?
This kind of search (called conjunction search) is slow if the surrounding objects are squares and other gray shapes
Conjunction searches are generally not preattentive

58. Coding with combination of features

59. Highlighting
The purpose of highlighting is to make something stand out from other information
As a rule of thumb, use whatever graphical dimension is least used otherwise in the design
If texture is not used elsewhere, use texture
Use motion (but may be too strong a cue)

60. Semantic depth of field
Use depth of field for highlighting

61. References
Jeremy M. Wolfe, How Might the Rules that Govern Visual Search Constrain the Design of Visual Displays? Society for Information Display Symposium Digest of Technical Papers, 36(1), pp
Colin Ware, “Information Visualization”, Morgan Kaufmann, 2nd edition, 2004.
Chapter 5 “Visual Attention and Information that Pops Out”