Representing Data using Static and Moving Patterns Colin Ware UNH
Introduction Finding patterns is key to information visualization. Expert knowledge is about understanding patterns (Flynn effect) Example Queries: We think by making pattern queries on the world Patterns showing groups? Patterns showing structure? When are patterns similar? How should we organize information on the screen? What makes a pattern distinct?
The dimensions of space
The “What” Channel Objects, any location Simple features specific locations Patterns of patterns
Patterns Feature heirarchy (learned) Contours and Regions (formed on the fly)
V1 processing Ware:Vislab:CCOM
Texture segmentation (regions)
Textures and low level features
Interference based on spatial frequency
Low level tuning based on feature maps
A diagram with same principle
Field, Hayes and Hess Contour finding mechanisms
Results rt = spl con cr br spl: Shortest path length con: continuity cr: crossings br: branches 1 crossing adds.65 sec 100 deg. adds 1.7 sec 1 crossing == 38 deg.
Connectedness Connectedness assumed in Continuity
Continuity Visual entities tend to be smooth and continuous
Continuity in Diagrams Connections using smooth lines
Ware:Vislab:CCOM LOC – generalized contour finding The mechanisms of line and contour
Closure Closed contours to show set relationship
Extending the Euler diagram
Collins bubble sets
More Contours Direct application to vector field display
How to add VS? Terminations Some End-Stopped neurons respond only with terminations in the receptive field. Asymmetry along path Halle’s “little stroaks” 1868
Modeling V1 and above Dan Pineo
Vector Field Visualization Laidlaw
Perceptually optimize for Some sub-set of task requirements An optimization process (NSF ITR) Identify a visualization Method and a paramaterization Streaklets: A generalized Flow vis technique Characterize solutions Define task requirements Advection path perceptio Magnitude perception Direction perception Human In the Loop Actual solutions Guidelines Algorithms Theory
Key idea Almost all solutions can be described as being composed of “streaklets” Mag color Mag luminance Mag size (length, width) Mag spacing Orient orient Direction arrow head Direction shape Direction lum change Direction transparency
Task: optimize streaklets. (How?) 1) Streaklet design optimized according to theory – head to tail, direction cues Modified Jobard and Lefer (Pete Mitchell) 2) Human in the loop optimization Genetic algorithms (NO) Domain experts with a lot of sliders Designers with a lot of sliders
Possibilities for Evaluation Direction Magnitude Advection Global pattern Local pattern Nodal points
Back to the feature hierarchy
Scatter plots: comparing variables
Parallel coords vs Generalized draftsmans plot
Parallel coord vs gen draftsmans Parallel Each line is a data Dimension Gen drafts All pairwise scatterplots. Results suggest Gen drafts is best Clusters & correlations Holten and van Wijk
Symmetry Symmetry create visual whole Prefer Symmetry
Symmetry (cont.) Using symmetry to show Similarities between time series data
Bivariate maps (texture + color)
3 Channels: Color, Texture, Motion
Compare to this!!
Scribble exercise
Ware:Vislab:CCOM The Magic of Line and Contour: Chameleon lines Saul Steinberg Santiago Coltrava
Ware:Vislab:CCOM
Patterns in Diagrams Patterns applied
Visual Grammar of diagrams Entities represented by Discrete objects Attributes: Shape Colors Textures Relationships represented by Connecting lines or nesting regions
Semantics of structure
Treemaps and hierarchies Treemaps use areas (size) SP tree Graph Trees use connectivity (structure)
Top down – Bottom up Tunable attention to patterns Contours and regions + Some are automatic Basic to constructive thinking
Part II: Patterns in Motion How can we use motion as a display technique? Gestalt principle of common fate
Motion as a visual attribute (Common fate) correlation between points: frequency, phase or amplitude Result: phase is most noticeable
Motion is Highly Contextual Group moving objects in hierarchical fashion.
Using Causality to display causality Michotte’s claim: direct perception of causality
A causal graph
Michotte’s Causality Perception
Visual Causal Vectors
Experiment Evaluate VCVs Symmetry about time of contact.
Results Perceived effect
Motion Patterns that attract attention (Lyn Bartram) Motion is a good attention getter in periphery The optimal pattern may be things that emerge, as opposed to simply move. We may be able to perceive large field patterns better when they are expressed through motion (untested)
Anthropomorphic Form from motion Pattern of moving dots (captured from actor body) – Johansson. Attach meaning to movements (Heider and Semmel)
Conclusion Gestalt Laws are useful as design guidelines. Patterns should be present in luminance Patterns should be the appropriate size Motion is under-researched, but evidence suggest its power. Simple motion coding can be used to express communication, causality, urgency, happiness? (Braitenberg)
Algorithms Optimizing trace density (poisson disk) Flexible methods for rendering (enhanced particle systems).
Figures and Grounds (cont.) Rubin’s Vase Competing recognition processes
Show particle solutions Problem: how do we create an optimal solution out of all of these possibilities? Standard solution: do studies and measure the effect of different parameters. Problem: Too many alternatives.
Closure (cont.) Segmenting screen Creating frame of reference Position of objects judged based on enclosing frame.
Laciness (Cavanaugh) Layered data: be careful with composites of textures
Transparency Continuity is important in transparency x y > z y z > w
Limitation due to Frame Rate Can only show motions that are limited by the Frame Rate. We can increase by using additional symbols.