510 likes | 611 Views
A Perceptual Heuristic for Shadow Computation in Photo-Realistic Images. Wednesday, 2 August 2006 Peter Vangorp Olivier Dumont Toon Lenaerts Philip Dutré Katholieke Universiteit Leuven. {peter.vangorp,olivier.dumont,toon.lenaerts,philip.dutre}@cs.kuleuven.be. Introduction.
E N D
A Perceptual Heuristic for Shadow Computation in Photo-Realistic Images Wednesday, 2 August 2006 Peter Vangorp Olivier Dumont Toon Lenaerts Philip Dutré Katholieke Universiteit Leuven {peter.vangorp,olivier.dumont,toon.lenaerts,philip.dutre}@cs.kuleuven.be
Introduction 3 types of realistic rendering
Introduction • 3 types of realistic rendering • radiometric accuracy“render everything a photometer can detect”
Introduction • 3 types of realistic rendering • radiometric accuracy“render everything a photometer can detect” • physiological perception“render only what the eye can see”
Introduction • 3 types of realistic rendering • radiometric accuracy“render everything a photometer can detect” • physiological perception“render only what the eye can see” • psychological perception“render only what the brain can see”
Radiometric accuracy measure shapes, light sources, materials, ... photometer Cornell box [Meyer et al. 1986]
Physiological perception use low-level limitations of human visual systemthreshold vs intensity, contrast sensitivity, ... reference threshold map visibly indistinguishable adaptive rendering [Ramasubramanian et al. 1999]
Physiological perception use low-level limitations of human visual systemthreshold vs intensity, contrast sensitivity, ... reference threshold map visibly indistinguishable adaptive rendering [Ramasubramanian et al. 1999]
Psychological perception use higher-level heuristics “Is a shadow necessary for therealism of a scene?” [Thompson et al. 1998]
Psychological perception • use higher-level heuristics “Is a shadow necessary for therealism of a scene?” “Do we need highlights to convey material properties?” [Thompson et al. 1998] [Fleming et al. 2004]
Psychological perception • use higher-level heuristics “Is a shadow necessary for therealism of a scene?” “Do we need highlights to convey material properties?” “How detailed should the geometry be?” [Thompson et al. 1998] [Fleming et al. 2004] [Luebke 2001]
Motivation & Goals Motivation shadows are important for perception of realism Goal detect perceptually important shadows in the scene render important shadows accurately approximate unimportant shadows [Kersten et al. 1997]
Methodology Psycho-physical experiments Derive a heuristic predicting shadow importance Design a perceptually driven algorithm Experimental validation
Methodology • Psycho-physical experiments • Derive a heuristic predicting shadow importance • Design a perceptually driven algorithm • Experimental validation
Methodology • Psycho-physical experiments • Derive a heuristic predicting shadow importance • Design a perceptually driven algorithm • Experimental validation
Methodology • Psycho-physical experiments • Derive a heuristic predicting shadow importance • Design a perceptually driven algorithm • Experimental validation
1. Perceptual experiments Test setup: 162 images, varying sphere over 9 radii and 9 heights correct shadow: no shadow: (avg. illumination)
1. Perceptual experiments • “Does the lighting in this image look realistic?” • single stimulus • 5000+ decisions • avg. 2 sec / decision
2. Heuristic large difference in realism, e.g. difference in realism small difference in realism, e.g. sphere height sphere radius
3. A perceptually driven algorithm • Ray tracing
3. A perceptually driven algorithm • Ray tracing • shoot viewing ray
distance solid angle 3. A perceptually driven algorithm • Ray tracing • shoot viewing ray • evaluate heuristic in hit point to be shaded in function of distance and solid angle
3. A perceptually driven algorithm • Preprocessing step • shadow photon map [Jensen and Christensen 1995]
3. A perceptually driven algorithm • Preprocessing step • shadow photon map • shadow photons augmented with heuristic [Jensen and Christensen 1995] shadow photon map
3. A perceptually driven algorithm For each viewing ray Gather nearest shadow photons Calculate average perceptual value yes no avg < threshold approximate: photon map render accurately: shadow rays
3. A perceptually driven algorithm For each viewing ray Gather nearest shadow photons Calculate average perceptual value yes no avg < threshold approximate: photon map render accurately: shadow rays
3. A perceptually driven algorithm For each viewing ray Gather nearest shadow photons Calculate average perceptual value yes no avg < threshold approximate: photon map render accurately: shadow rays
3. A perceptually driven algorithm For each viewing ray Gather nearest shadow photons user-defined threshold Calculate average perceptual value yes no avg < threshold approximate: photon map render accurately: shadow rays
3. A perceptually driven algorithm For each viewing ray Gather nearest shadow photons user-defined threshold Calculate average perceptual value yes no avg < threshold approximate: photon map render accurately: shadow rays
3. A perceptually driven algorithm For each viewing ray Gather nearest shadow photons user-defined threshold Calculate average perceptual value yes no avg < threshold approximate: photon map render accurately: shadow rays
4. Validation Similar perceptual experiment “Do the lighting and the shadows look realistic?”
4. Validation • Similar perceptual experiment • “Do the lighting and the shadows look realistic?” • Stimuli: 6 scenes • threshold 25%, 50%, 75% • reference rendering (threshold 0%)
4. Validation • Similar perceptual experiment • “Do the lighting and the shadows look realistic?” • Stimuli: 6 scenes • threshold 25%, 50%, 75% • reference rendering (threshold 0%) • 15 subjects, almost 6000 decisions, avg. 5 seconds
4. Validation Up to 50% of the pixels can be approximated, without loss of perceptual realism Threshold 25% Approx px 13% Threshold 50% Approx px 24% Threshold 75% Approx px 48%
Results threshold 90% reference
Results threshold 90% reference
Results threshold 90% reference
Results threshold 90% reference
Results threshold 80% reference
Results threshold 80% reference
Results threshold 80% reference
Results threshold 80% reference
Results threshold 90% reference
Results threshold 90% reference
Results threshold 90% reference
Conclusions & Future Work Intuitions confirmed by statistical data Rendering algorithm driven by perceptual information
Conclusions & Future Work • Intuitions confirmed by statistical data • Rendering algorithm driven by perceptual information • Extend methodology to other phenomena • Different questions than “Does this look realistic?” • Better ways to incorporate perceptual information • currently no significant acceleration yet