Related Experiment Videos
Statistical characterization of real-world illumination
Ron O Dror1, Alan S Willsky, Edward H Adelson
1Massachusetts Institute of Technology, Cambridge, MA, USA. rondror@ai.mit.edu
Journal of Vision
|October 21, 2004
Summary
Real-world illumination, captured in spherical illumination maps, shows surprising statistical regularity. This finding advances computer vision and realistic graphics rendering by modeling complex lighting.
Area of Science:
- Computer Vision
- Computer Graphics
- Human Perception
Background:
- Traditional vision and graphics studies use simplified illumination models.
- Real-world lighting is complex, with light incident from many directions.
- Spherical illumination maps capture omnidirectional lighting at a point.
Purpose of the Study:
- To analyze the statistical properties of real-world illumination.
- To establish a foundation for statistical illumination models.
- To improve understanding of material perception, computer vision, and graphics rendering.
Main Methods:
- Acquisition of high dynamic range (HDR) spherical illumination maps.
- Statistical analysis of wavelet coefficient distributions.
- Analysis of harmonic spectra of illumination maps.
Main Results:
- Real-world illumination exhibits significant statistical regularity.
- Illumination maps share similarities with natural image statistics (wavelet and spectral properties).
- Illumination maps are statistically non-stationary and can feature dominant localized light sources.
Conclusions:
- Statistical models of real-world illumination are feasible.
- Findings support enhanced material perception, robust computer vision systems, and realistic computer graphics.
- Understanding illumination statistics is key for advancing visual computing fields.