natural visualization steve haroz & kwan-liu ma university of california at davis
TRANSCRIPT
Natural Visualization
Steve Haroz & Kwan-Liu Ma
University of California at Davis
Outline
• Purpose
• Math Background
• Applying and extending existing theories
• InfoVis Contest
• Application to GUIs
• Summary and Conclusion
Outline
• Purpose
• Math Background
• Applying and extending existing theories
• InfoVis Contest
• Application to GUIs
• Summary and Conclusion
Purpose
• What makes for a good Visualization?– Aesthetics?– Color?– Complexity?– Beginner or Expert? Intuitive?
• Can understanding the process of visualization help?
The Process
0 1 0 1 1
1 0 1 1 0
1 0 0 0 1
…
Visualization
Complete?
Which Representation Is Best?
“Who can prove by experience the non-existence of a cause when all that experience tells us is that we do not perceive it?”
The Process
0 1 0 1 1
1 0 1 1 0
1 0 0 0 1
…
Visualization
Hubel 1988
The Forgotten Stage of Visualization
Purpose
• Applicability of visual system knowledge– Retina “tuned” to natural images
• Certain images more easily perceptible?
• Is interaction aided by these “natural GUIs” ?
Outline
• Purpose
• Math Background
• Applying and extending existing theories
• InfoVis Contest
• Application to GUIs
• Summary and Conclusion
Spatial Frequencies
• Similar to auditory frequencies
• Varying intensity (light) over space
Fourier Transform
• Sum of sin/cos waves
Spatial Frequencies of Natural Images
• Take Fourier transform along each orientation and average
• f -2 pattern
• Pattern is prevalent in all natural scenes
• Plot on log-log scale
Unnatural images
Natural Images
Size Distribution
• This pattern is explained bya ‘collage’ of objects occluding each other
• These objects have a power distributionarea = 2x
Outline
• Purpose
• Math Background
• Applying and extending existing theories
• InfoVis Contest
• Application to GUIs
• Summary and Conclusion
power exponential
linear constant
Plot of spatial frequencies
Const
Exponential
Linear
Power
Linear Trend
-2.67
-2.66
-2.65
-2.64
-2.63
-2.62
-2.61
-2.6
-2.59
Constant Linear Exponential Power
Images Without Occlusion
You can’t visualize what is not visible
• Images with adjacent squares
• Same sizing applies
power exponential
linear constant
Trend – no occlusion
-2.79
-2.77
-2.75
-2.73
-2.71
-2.69
-2.67
Constant Linear Exponential Power
Outline
• Purpose
• Math Background
• Applying and extending existing theories
• InfoVis Contest
• Application to GUIs
• Summary and Conclusion
Naturalness Metric
1. Closeness to f-2
2. Linearity
InfoVis 2004 Contest
0
0.1
0.2
0.3
0.4
0.5
1st Place 2nd Place
InfoVis 2005 Contest
0
0.1
0.2
0.3
0.4
0.5
0.6
1st Place 2nd Place 3rd Place
Outline
• Purpose
• Math Background
• Applying and extending existing theories
• InfoVis Contest
• Application to GUIs
• Summary and Conclusion
Image Analysis for GUI Study
• Applications with hierarchical data
• Analyze screenshots
• Compare with usage data (user study)
• Use statistics to find behavioral patterns
Correlation with Response Time
0
0.2
0.4
0.6
0.8
1
1.2
1 2 3 4 5 6
|Slope+2||Avg Dev|
Outline
• Purpose
• Math Background
• Applying and extending existing theories
• InfoVis Contest
• Application to GUIs
• Summary and Conclusion
Summary and Conclusion
• Visualization preference correlates with a property of the visual system
• Bias-free metric may help vis generation
• Utility or aesthetics?
• More visual properties
Acknowledgements
• Bruno Olshausen
• Yue Wang