Transcript pattern perception - UTRGV Faculty Web

Static and Moving Patterns
Ware Chapter 6
Introduction
•
Title of Ware’s chapter: “Static and Moving Patterns”
•
Finding patterns, a kind of “insight” is key to visualization!
•
Investigation, exploration, discovery, … is often about finding patterns
–
–
–
–
That were previously unknown, or
That depart from the norm.
Finding such patterns can lead to key insights
• One of the most compelling reasons for visualization
• Computing is about insight, not numbers, Hamming, 1973
Goals of insight:
• Discovery
• Decision making
• Explanation
•
Expert knowledge is about understanding patterns
•
Example Questions:
–
–
–
–
Patterns showing groups?
Patterns showing structure?
When are patterns similar?
How should we organize information on the screen?
Some memories …
• From the first class …
Visualization – Main Ideas
• Definition:
– “The use of computer-supported, interactive visual representations of
data to amplify cognition.”
• Card, Mackinlay Shneiderman ’98
• This is among the most widely accepted contemporary working definitions
• Visuals help us think
– Provide a frame of reference, a temporary storage area
• Cognition → Perception
• Pattern matching
• External cognition aid
– Role of external world in thinking and reason
• Larkin & Simon ’87
• Card, Mackinlay, Shneiderman ‘98
When to use Visualization?
• Many other techniques for data analysis
– Data mining, DB queries, machine learning…
• Visualization most useful in exploratory data
analysis:
– Don’t know (exactly) what you’re looking for …
– Don’t have a priori questions ...
– Want to know what questions to ask
Data Analysis and Logical Analysis
• Data Analysis
– Data in visualization:
• From mathematical models or computations
• From human or machine collection
– Purpose:
• All data collected are (should be) linked to a specific relationship or theory
• Relationships are detected as patterns in the data
– Maybe call it insight
– Relationship may either be functional (good) or coincidental (bad)
– Data analysis and interpretation are functionally subjective
• Logical Analysis
– Applying logic to observations (data) creates conclusions (Aristotle)
– Conclusions lead to knowledge (at this point data become information)
– There are two fundamental approaches to generate conclusions:
• Induction and Deduction
• Equally “real” and necessary
Mueller, 2003
Deduction vs. Induction
• Deductive logical analysis probably the
more familiar
– Presented in detail since middle school
• Formulate a hypothesis first, then test
hypothesis
– via experiment and accept/reject
– data collection more “targeted” than in
induction
• i.e., only addressing “truth” (actually
falseness) of hypothesis
– only limited data mining opportunities
Mueller, 2003
Deduction vs. Induction
•
Inductive logical analysis part of scientific process,
and reasoning generally, but perhaps delineation if
its elements less familiar
•
Like, where do the hypotheses come from?
– Insight?
•
Make observations first, then draw conclusions
– organized data survey (structured analysis,
visualization) of the raw data provide the basis for the
interpretation process
– interpretation process will produce knowledge that is
being sought
– experience of individual scientist (observer) is crucial
– important: selection of relevant data, collection
method, and analysis method
– data mining is an important knowledge discovery
strategy
– ubiquitious data collection, filtering, classification, and
focusing is crucial
Mueller, 2003
Ware refocus …
• Ware notes that much of visualization is about “finding patterns”
– And Ch. 6 “Static and Moving Patterns” starts new part of book in
which low level parallel processing of features
– … Is combined with other “higher level” processes for “deeper”
analysis
• Will examine the role of pattern perception
– Mostly 2d processes occurring between feature analysis and full object
perception
– “flexible middle ground where objects are extracted from patterns of features”
– Point where “bottom up” feature extraction meets “top down” object recognition
processes
• “Understanding pattern perception provides abstract design rules
that can tell us much about how we should organize data so that
important structures will be perceived”
Recall … Model of Perceptual Processing
What we do is design information displays!
•
An information processing (the dominant paradigm) model
–
“Information” is transformed and processed
•
–
Gives account to examine aspects important to visualization
–
In spirit of visualization as evolving discipline, yet to develop its theories, laws, …
•
•
•
Here, clearly, many neural subsystems and mapping of neural to ip is pragmatic
Stage 1: Parallel processing to extract low-level properties of the visual science
Stage 2: Pattern perception
–
•
Physical light does excite neurons, but at this “level of analysis” consider information
Today, focus on first elements of this stage, things that “pop out”
Stage 3: Sequential goal-directed processing
Stage 2: Pattern Perception
•
Rapid, “active”, but not conscious processes
•
Specialized for object recognition
–
–
•
Specialized for interacting with environment
–
•
Visual attention and memory
• E.g., for recognition must match features
with memory
Task performing will influence what perceived
• Bottom up nature of Stage 1, influenced
by top down nature of Stage 3
E.g., tasks involving eye-hand coordination
“Two-visual system hypothesis”
–
One system for locomotion and eye-hand
coordination
•
–
One system for symbolic object manipulation
•
•
The “action system”
The “what system”
Both bottom up and top down
–
More emphasis on arbitrary aspects of symbols
than Stage 1
Detail: Model of Object Perception
•
Stage 1: Parallel, fast extraction
–
–
•
Stage 2: Pattern Perception
–
–
•
Contours and boundaries form perceptually distinct
regions
Will study this “middle ground” today
Stage 3: Object Identification
–
–
•
Form, motion, texture, color, stereo depth
Contrast sensitivity, edge detection, as before
Slower, serial identification of objects within scene
Comparisons with working memory
There is feedback
–
–
Linear model is a simplification
Later stage intentions affect earlier stage
responses
Parallel feature processing:
orientation, texture,
color, motion. …
Detection of 2D patterns,
contours, regions, …
Object identification,
working memory, …
To Learn about Pattern Perception…
• Will examine:
–
–
–
–
–
Gestalt laws
Contour perception
Perception of transparency: Overlapping data
Perceptual syntax of diagrams
Patterns in motion
• Perception of Causality
So, what’s this?
• x
Perception - Emergence
• Local (small) regions of image not
contain sufficient information to
extract contours from noisy edges
– E.g., front paw
• Upon recognition of form (dog)
contours are perceptually evident
• But, there are no contours in
senses we have discussed so far!
– Rather, it is the filling in of voids
that leads to contours
– Clearly contours “constructed”
• Dog perceived as a “whole”
• And the perceived image (whole) is
“more than the sum of its parts”
Perception - Multistabilty
• Necker cube
– Which is the front face?
• Shows multistability of perception
– Perception not a sequential process from input to percept,
– Is dynamic system - equilibrium state(s) represent final percept
Perception - Constructive
• Object perceived as having
more spatial information than
is actually present in image
• Again, perception “fills in the
blanks”
• Subjective contour illusions
illustrate
• In perception, tend to order
our experience in a manner
that is regular, orderly,
symmetric, and simple
• Gestalt
Gestalt Laws – Perception, Organization
• First attempt to understand pattern
perception
• 1912, “Gestalt school of psychology”
– Max Wertheimer, Kurt Koffka, and
Wolfgang Kohler, 1930’s
• Gestalt = pattern/form (in German)
Max Wertheimer, 1880-1943
Kurt Koffka, 1886-1941
• Perceptual organizing principles
• Patterns transcend visual stimuli that
produced them
• Got “laws”, or rules of pattern
perception, essentially right, if not
mechanisms
– “Laws” still hold, different explantions
Wolfgang Kohler, 1887-1967
Gestalt Laws – perception, organization
• Robust “laws” easily translate into
design principles:
– Figure and Ground
– Proximity
– Similarity
– Continuity (and Connectedness)
– Symmetry
– Closure
– Relative Size
– Common Fate (motion perception)
Max Wertheimer, 1880-1943
Kurt Koffka, 1886-1941
Wolfgang Kohler, 1887-1967
Gestalt Law: Figure and Ground
• What is foreground, what is
background?
– At right is there a black object on
a white background, or
– A white object on a black
background?
• Fundamental perceptual act in
object identification according to
Gestalt school
• All other principles help determine
this
• Symmetry, white space, and
closed contours contribute to
perception of the figure
Gestalt Law: Figure and Ground
• Rubin’s Vase
– Another “illusion”
• What is figure what
is ground?
• Can they swap?
– Suggests
Competing
recognition
processes
– Following slides
illustrate
1
• Again
2
• And again
3
•
And again
4
• And again
One Last Figure Ground Example
• A man playing saxophone or a woman’s head?
Gestalt Law: (Spatial) Proximity
a
• Principle: Things close (physically) are grouped
together
– One of most powerful perceptual organizing principles
• Spatial concentration principle –
– Perceptually group regions of similar element density
•
We perceptually group regions of similar density
– “Perceptually” means without conscious intervention
– It is as if the “groupings” are inherent in environment
• To a larger extent than they are, recall edges
• E.g., Below dots clearly perceived as rows and
columns, though difference in spacing is small
x
b
Gestalt Law: Similarity
• Principle: Things that are “similar”, by some criterion, are grouped
together
• Again, “perceptually” …
• Visual groupings by similarity
• Below, color or shape similarity groups by row
a
b
Similarity: Integral and Separable Dimensions
• Color or shape similarity groups by
row
a
b
• Separable dimensions enable
alternate perception
– E.g., in 6.5 integral dimensions on left,
separable on right
Separable dimensions
Integral dimensions
Integrable dimensions form stronger pattern
Gestalt Law: Connectedness
• Principle: Connectedness (association, grouping) can be shown
explicitly
– Stronger than proximity (a), color (b), size (c), and shape (d)
– Assumed in Continuity
– Connecting different graphical objects by lines is a powerful way of
expressing that there is some relationship among them
• E.g., node-link diagrams
a
c
b
d
Gestalt Law: Continuity
•
Principle: More likely to construct
visual entities out of visual
elements that are smooth and
continuous, rather than ones that
contain abrupt changes of
direction
– As shown in examples a-c (top)
•
Visual entities tend to be smooth
and continuous
– “Good continuity” of elements
•
Connections using smooth lines
facilitate perception continuity, as
shown in a, b (below)
a
b
Gestalt Law: Symmetry
• Principle: Symmetry creates
visual whole
– Bilateral symmetry stronger
than parallelism
• Prefer symmetry
– Symmetric shapes seen as
more likely
– Explains why cross shape so
clearly perceived – vs. b
• Make use of symmetry to
enable user to extract similarity
(next slide)
Gestalt Law: Symmetry
• Design principle:
Make use of
symmetry to
enable user to
extract similarity
– Ex.
Gestalt Law: Closure
• Principle: A closed contour is seen as an object
• Perceptual system will close gaps in contours
– System “prefers” closed contours
– E.g., tend to see a as a circle obscured by rectangle
• Rather than a circle with a gap by a rectangle
• Word “closure” has entered language with variety of meanings
a
b
Gestalt Law: Closure
• Contour separates world into “inside” and
“outside”
– Stronger than proximity
– Venn diagrams from set theory
• Closed contours to show set relationship
– Closure and continuity both help
• Closed rectangles strongly segment visual
field
– Provide frames of reference
– Position of object judged based on enclosing
frame
• Design Principle:
– Partial obscuration is okay
– Especially for symmetric objects
A
B
C
D
Gestalt Law: Closure
• Ware: Extending Venn diagram
– Adding color, texture, etc. facilitates “closure” and contour
perception
Gestalt Law: Relative Size
• Principle: Smaller components of
a pattern tend to be perceived as
the object
– E.g., black propeller on white
background
• Horizontal and vertical tend to be
seen as objects
• Plays into figure/ground principle
• Design principle
– Make dots the objects, rather than
elements of a figure, e.g., “cheese
grater”
Contours
• Contour: A Perceived continuous
boundary between regions
• Can be defined by:
– Line (sharp change on both sides
in intensity)
– Boundary between regions of two
colors
– Stereoscopic depth
– Patterns of motion
– Texture
– Even perceived where are none
• E.g., illusory contour at right
– Boundary of blobby shape
• continuity & closure
FYI: Perceiving Direction: Representing Vector
Fields
• How to represent vector
(direction) fields?
a
– Frequent in scientific
visualization
– Need to represent:
• Orientation
• Magnitude
b
• When do contours jump gaps?
– When a smooth curve can be
drawn over gaps.
• E.g., at right b (shifted next) is
easier to see flow in that a
– Straight lines are easiest
– Quite wiggly is possible
– Direct application to vector
field display
Perceiving Direction: Representing Vector Fields
• Which direction?
• Away from background
(Opt.) 2D Flow Visualization Techniques
•
An experimental
comparison
–
–
–
•
But far from exhaustive
Cf. Ware
(labeled A-C 1st row, DF 2nd row)
Figures:
A. Arrows on a regular grid –
fixed length
B. “ jittered grid C. Triangle icons – size
proportional to field strength
D. Line integral convolution
E. Large head arrows along
a streamline, regular grid
F.
“
, constant spacing
•
Tasks:
–
–
0 magnitude, hardest
with a & b
Trajectories, f best, d
worst
(Opt.) 2d Flow Visualization
• .
Kirby, R.M., M.M. and D.H. Laidlaw. Visualizing
Multivalued Data from 2D Incompressible
Flows Using Concepts from Painting, IEEE
Visualization 99, San Francisco, CA, IEEE
Press, pp. 333-340, 1999.
(Opt.) 2d Flow Visualization
•
Visual Thinking
for Design, Ware
(opt.) Perception of Transparency: Overlapping Data
•
Presentation of data in “layered” form common
visualization technique
– E.g., GIS
– Common technique is to present one layer of data as
if transparent layer over another
– But, problems –
x
• contents of layers always interfere to some extent with
others
• Sometimes layers will fuse perceptually, so not possible
to determine to which layer object belongs
b
a
•
Main determinants of perceived continuity:
– Good continuity (a), and
– Ratios of gray values (or colors) in different pattern
elements (a)
•
x
Rules for transparency to be perceived
w
y
– Where x, y, z, w are gray values
– x < y < z or x > y > z
– y < z < w or y > z > w
b
a
z
(opt.) Perception of Transparency: Overlapping Data
•
Also, can represent layers of data by showing
each as a see-through texture or screen pattern
a
•
–
–
–
a & b, clear layers
c not
d bistable,
•
•
•
Be careful with composites of texture
c
Many perceptual pitfalls
–
–
–
•
sometimes two different
Sometimes three, with -, | , and + elements
Attempting to present multiple data layers
–
•
b
Experiment with right shows possibilities tested
Different layers interfere with each other to some
extent
Sometimes layers will fuse perceptually into one
Patterns similar in color, frequency, motion, etc.
interfere more
Design principle:
–
Make layers differ in at least one significant
dimension
d
(Opt.) Perceptual Syntax of Diagrams
• Diagrams are hybrids of:
– Conventional (learned) elements
• E.g., labeling codes such as math symbols
– Perceptual elements
• E.g., as shown in Gestalt laws, grouping
• Graphs are natural and ubiquitous form of
information display
– “graph drawing algorithms”, subject of study in computer
science
• Quantitatively describing “good layout” is challenge
– Nodes represent something,links represent something else
• E.g., entity-relation diagrams
– Can consider a graph a diagram
• Ware suggests a “grammar of node-link diagrams”
– Way of describing elts of diagrammatic use of
graphs
– I.e., to convey information
•
Four kinds of node-link diagrams used in software engineering
–
–
(with text labels on nodes and arcs)
A. code module diagram, B. data flow diagram, C. object modeling diagram, D. state transition
diagram
(opt.) Grammar of Node-Link Diagrams
• Table expresses ways in which entities and relationships can be
expressed using node-link diagrams
• Visual grammar:
– “Standardization” (or agreement) of interpretation of visual elements
Graphical code
Visual instantiation
Semantics
Graphical code
1. Closed contour
Entity, object, node
8. Spatially ordered shapes
A sequence
9. Linking line
Relationship between
2. Shape of closed region
Entity type
3. Color of enclosed region
Entity type
4. Size of enclosed region
Entity value
Larger = more
5. Partitioning lines within
Entitity partitions are
enclosed region
created
e.g. treemaps.
6. Attached shapes
contour
Semantics
entities
10. Linking line quality
Type of relationship
between entity
11. Linking line thickness
Strength of relationship
between entities
12. Tab connector
A fit between components
13. Proximity
Groups of components
Attached entities
Part_of relations
7. Shapes enclosed by
Visual instantiation
Contained entities
(opt.) Visual Grammar of Map Elements
• Similarly, a visual grammar exists for maps
• Only three basic kinds of graphical marks are common to most
maps:
– Areas, line features, point features
Graphical code
Semantics
Graphical code
1. Closed contour
Geographic region
7. Dot in closed contour
2. Colored region
Geographic region
3. Textured region
Geographic region
4. Line
Linear map features such as
5. Dot
6. Dot on line
Visual Instantiation
Visual Instantiation
Semantics
Point feature such as town
located within a geographic
region.
8. Line crosses closed
Linear feature such as river,
contour
crossing geographic region.
rivers, roads, etc. Depends
region
on scale.
9. Line exits closed
A linear feature such as a
Point features such as town,
contour
river terminates in a
building. Depends on scale
region
geographic region.
Point feature such as town
10. Overlapping contour,
Overlapping geographically
on linear feature such as
colored regions,
Defined areas.
road.
textured regions.
Use in Design - Example
•
Visual Thinking for Design, Ware
Patterns and Attention
•
“Visual Thinking for Design, Ware
Patterns in Motion
• So far, have discussed static patterns only
– By far largest number of visualizations are static
– Perception of dynamic patterns less well understood
• Still, humans very sensitive to:
– Motion generally
– Patterns in motions
• Will see examples and overviews – causality, why wagon wheel effect
– Gestalt principle of “common fate”
• But, it is complicated, and we’ll only touch on it
• Next is an example
Albert Michotte’s Motion Experiments, 1946
• Humans find order …
• Humans ascribe causality …
• Michotte’s demonstrations …
– 4 + conclusion
• http://cogweb.ucla.edu/Discourse/Narrative/michotte-demo.swf
– Prepared by Warren Thorngate, Professor
Perception of Causality
• Michotte’s claim: Direct perception of
causality
• Task:
– Vary time delay from time dot moves
– Chart shows subjects’ perception of
causal relationship
100%
Direct Launching
Delayed launching
No causality
50%
100
Time (msec.)
200
Another Example
• Another example
– http://www.michaelbach.de/ot/mot_wagonWheel/index.html
– https://www.youtube.com/watch?v=e_jKNlC2YKo
Patterns in Motion, Example 2, 1
• How to represent data communication with animation?
– Let graphical object represent each “data packet”
– Then animate that package from information source to destination
• Observe bottlenecks, bursts, etc.
• For animation, with observer perceiving smooth motion, need to:
–
–
–
–
Display element at point a
Display element at point a’
Repeat (at a pretty fast frame rate)
Sequence of static pictures is then perceived as smoothly moving object

a
• But, limitation on “throughput”,
– i.e., how much data can be displayed per unit time
– Here, amount that an object can be moved before it becomes confused with
another object in the next frame
•
E.g., next spoke of the wagon wheel
– Correspondence problem
b
Patterns in Motion, Example 2, 2
• … limitation on “throughput” …
correspondence problem
– how much data can be displayed per unit time
– Amount that an obj can be moved before
confused with another obj in next frame
• Let  = distance between pattern elements
– The distance at which subsequent display of
elements is “right on top of” the next
  / 2 (in practice minus a bit, /3 empirically) is
maximum displacement/inter-frame movement
for element before the pattern is more likely to
be seen as moving in reverse direction than
what intended
• When elements identical, brain constructs
correspondences based on object
proximity in successive frames
– “wagon-wheel” effect
– With /3, frame rate = 60 fps, have upper bound
of 20 messages per second

a
b
c
Form and Contour in Motion
• Might use motion to code attributes, etc.
• Patterns of dots moving in synchrony group together
– Gestalt principle of “common fate
•
Demo http://tepserver.ucsd.edu/~jlevin/gp/time-example-common-fate/
• Contours seen in moving dot fields by motion alone
– Rivals static contour detection
• Phase of the motion seems most salient
– Compared to frequency and amplitude
a
• Design Principle:
– Consider animation for association of groups
• Might also, group moving objects in hierarchical
fashion
– Moving frames, next slide
b
Moving Frames
• Rectangular frame forms strong context
– The stationary Dot is perceived as moving in (a).
– The circle has no effect on this process in (b).
a
b
• Groups of dots moving together form frame
a
b
More About Motion
• Motion Design Principles:
– Use motion as strong cue for grouping
– Add frame around group of related particles
– Speed around a few cm per second
• Speed up things that are much slower than this
• Slow down things that are much faster
• Other Motion Information
– Motion can express causality
– Motion of dots on human limbs immediately
recognizable
– Motion patterns can express emotion or behavior
Perception of Animate Motion
• Pattern of moving dots (captured from actor body) – Johansson
– People can identify anger, gender, etc.
– Today, motion capture at heart of animation development
– http://www.mocapdata.com/
• Attach meaning to movements (Heider and Semmel)
a
b
End
• .