Psychophysical Image Quality Metrics

1
Psychophysical Image Quality Metrics

• Ann McNamara
• Image Synthesis Group
• Trinity College, Dublin, Ireland

2
Psychophysical Image Quality - Overview

• Realism
• Psychophysics
• Working with Participants
• Procedures for comparing Real Scenes
• with Synthetic Image
• Case Study A Psychophysical Investigation

3
Need For Accuracy

• Lighting Engineering
• Architecture
• Stage Lighting
• Industry
• Entertainment
• Safety Critical
• Archaeology

4
Why Compare Images ?

• Compare and validate lighting simulations
• Use comparisons to guide rendering
• more efficiently
• Can we compute less without altering human
  perception of an image
• While pixel by pixel comparison might be 0,
  human might not see any difference

5
Pixel by Pixel Comparison
6
Visual Psychophysics

• Determine the relationship between the
  physical world and humans subjective experience
  of that world

7
Visual Psychophysics

• Measure the mind
• Without bias
• Systematically
• Repeat Observations
• Relationship between mind matter

8
Experimental Design

• Make inferences without ambiguity
• Rule out alternative causes, leaving only
• the actual factor that is the real cause

9
Experimental Design
Plan
10
Experimental Design

• Selecting what to study
• Selecting who to study
• Specifying how to study
• Specifying the sequence of
• measurements to be recorded
• What kind of evidence will result

11
Order Effects

• Order of Presentation
• Good before Bad
• Timing

12
Randomisation

• Not haphazard
• No event is ever predictable from
• any of the preceding sequence

13
Counterbalancing

• Order of Presentation
• Influence Results
• Make Experiment fair
• Reduce Bias

14
Control of Extraneous Variables

• May influence or affect the results of
• the condition
• For Example - Outside illumination

15
Experimental Design

• The Question to be Answered
• Choice of Task (Measure)
• Choice Control of Physical Stimuli
• Organisation of Participants
• Sequence of Presentation
• Instructions to Participants
• Recording, Presenting Analysing Results

16
Psychophysics for Judging Image Quality

• Psychophysical methods allow us to ask
• how close to reality computer images are
• Validate progressive global illumination
• solutions

17
Case StudyComparing a real scene with computer
generated images that represent that scene
18
A method of comparing real scenes with graphics

• Task Estimate the lightness of
• various regions of a scene
• Lightness perception is known
• to depend on prior perception of 3-D
• shape and illumination

19
A method of comparing real scenes with graphics

• Lightness is therefore a useful measure
• of the fidelity of illumination and
• 3-D reconstruction of a graphics scene

20
Why Lightness ?
21
Physical Stimulii
22
Physical StimuliThe Real Scene

• Painted 5-sided Cube
• Objects painted with different grey paints
• Complex illumination, with secondary
  reflections

23
Physical StimuliGraphic Reconstructions
24
Participants and Sequencing

• How Many Participants
• Randomisation of Participants
• Randomisation of presentation
• Time of Day Influence
• Training on Physical Stimuli

25
Physical StimuliTraining on Patches
26
Physical StimuliTraining on Patches
27
Experiment
28
Case Study Results

• Average match in each image is plotted
• along side average match in the real scene

29
Real Photograph
30
Real Two Ambient Bounce
31
Real Eight Ambient Bounce
32
Real Eight Ambient Bounce Bright
33
Real Default
34
Real Estimated Materials
35
Real Estimated Illumination
36
Real Tone Mapped
37
Real Raytraced
38
Real Radiosity
39
Data Analysis

• Correlation - indication of how closely related
  two sets of data are
• ANOVA - Analysis Of Variance
• T Tests
• Statistics reinforce the evidence
• from the data

40
Summary

• Experiments should be designed to
• produce accurate results
• Attention must be paid to a number of
• subtle experimental issues such as sample
• size, bias and randomisation

41
Further Information

• HTTP//www.cs.tcd.ie/Ann.McNamara
• HTTP//www.isg.cs.tcd.ie/Campfire
• Birds of a Feather Perception and Graphics,
  Monday 13th August, 1-2pm, Room 506

42
Psychophysical Image Quality Metrics

• Ann McNamara
• Image Synthesis Group
• Trinity College, Dublin, Ireland