Parameter Estimation for
Photographic Tone Reproduction

Abstract

Tone reproduction algorithms attempt to compress a large range of pixel values into a smaller range that is suitable for display on devices with limited dynamic range. A recently proposed tone reproduction operator achieved this goal by emulating photographic practice. Like in photography, the manual tuning of a small set of intuitive parameters was required. In this paper we extend this photographic tone reproduction algorithm with an automatic method for estimating these parameters. The estimation process is also based on photographic practice. The resulting operator produces good images and does not require manual parameter tuning. Sample source code is available online.

Paper

PDF

Source code

Download the Source code.

This code runs on SGI workstations without modifications. It also runs on Sun workstations, using the FFTW public domain fft library. This set-up is likely to work on other flavors of Unix, but has not been tested on other platforms.

Results

Gamma 1.6	Gamma 2.2
bathroom	bathroom
Cornell box	Cornell box
Desk	Desk
Fog	Fog
GroveC	GroveC
GroveD	GroveD
Sun	Sun
Lamp	Lamp
Memorial	Memorial
Nave	Nave
Parking garage	Parking garage
Rosette	Rosette
Vinesunset	Vinesunset

Source images are courtesy of Paul Debevec, Sumant Pattanaik, Peter Shirley, Jack Tumblin and Greg Ward

Links

Previous work on photographic tone reproduction.
High dynamic range data.

Acknowledgments

Many researchers have made their high dynamic range images and/or their tone mapping software available, and without that help our comparisons would have been impossible. Thanks also to Greg Ward for his Radiance read and write functions. This work was supported by NSF grants 89-20219, 95-23483, 97-96136, 97-31859, 98-18344, 99-77218, 99-78099, EIA-8920219 and by the DOE AVTC/VIEWS.