Retinex Workshop 8:00 – 12:30

Chairs: Marcelo Bertalmío, University Pompeu Fabra (Spain); Alessandro Rizzi, University of Milano (Italy); and John McCann, McCann Imaging (USA)

8:00 AM Introductions of Workshop Participants: Organizers and Participants

John McCann

  1. Introduction
  2. Retinex Theory of Color Vision
  3. Three independent spatial channels
  4. Color correlates with three spatial lightnesses
  5. Tests of Retinex’s ability to predict color appearances
  6. Discussion
  1. Modeling Lightness – The spatial calculation for each Retinex channel
  2. Two different goals of Lightness calculations
  3. Model of human vision –
  4. Tested by the ability to predict observer color matches
  5. Improve color photography’s dynamic range
  6. Tested by preferred scene rendition
  7. Study both in parallel
  8. Use successful models of vision to make better pictures
  9. Calculate appearance match and add preferred image enhancements
  10. Ratio-threshold-product-reset models
  1. Tuning spatial interaction parameters to accurately calculate visual sensations
  2. Many different types of target (with careful radiance calibration)
  3. Measurements of gradient threshold - Laplacians (Horn, Marr)
  4. Simultaneous Contrast was the most discerning test
  5. Visual contrast is neither local, nor global
  6. Discussion
  7. Summary and Take-Home messages

Alessandro Rizzi

  1. Milan Retinexes
  2. the problem of locality
  3. Paths
  4. BRW
  5. Montagna - Finlaynson
  6. Termite
  7. Sprays
  8. RSR
  9. ACE
  10. RACE
  11. STRESS
  1. Discussion

Break (Halfway)

Alessandro Rizzi

  1. Land’s Designator
  2. Grayworld
  3. No reset
  4. Hurlbert, Moore
  5. NASA Retinex
  6. Single scale
  7. Multi-scale
  8. Kotera
  9. Meylan and Süsstrunk
  10. Ramponi
  11. Discussion

Marcelo Bertalmío

  1. Perceptual color correction through variational techniques: linking local histogram equalization to color constancy and contrast enhancement.
  2. A partial differential equation formulation for Retinex exposes the connections between Retinex, efficient coding and neural models for visual activity.
  3. Accounting for induction (assimilation and contrast).
  4. Applications: color constancy, contrast enhancement, haze removal, tone mapping of high dynamic range images, gamut reduction and gamut extension for cinema, image fusion of exposure bracketing pictures.

Summary

The Retinex theory of Edwin Land proposed a model of human color vision based on color perception experiments. The original Retinex algorithm, by Land and McCann, is capable of performing color correction and contrast enhancement of images, with some limitations regarding overexposed pictures and visual artifacts stemming from its implementation based on one dimensional paths. By replacing these paths by two dimensional kernels while maintaining all the basic tenets of the theory, another Retinex algorithm was obtained which does not produce artifacts and that can be extended into a variational formulation allowing to handle naturally both over and underexposed images. Furthermore, this variational formulation linked Retinex with histogram equalization in image processing and with efficiency of representation and lightness induction in visual neuroscience. This workshop aims to provide the attendants with a novel look at the Retinex theory of color vision, seeing its contributions in a different light and showing the potential for applications to problems in image processing and computer graphics such as color constancy, contrast enhancement, haze removal, tone mapping of high dynamic range images, gamut reduction and gamut extension for cinema, and color transfer.