Vignette and Exposure Calibration and Compensation

Dan B Goldman         Jiun-Hung Chen
University of Washington

All photographic images have some degree of vignetting, a gradual darkening of the image toward its periphery. We have developed a technique to calibrate and remove vignetting, using only a sequence of images such as those used to create a panoramic mosaic. We can also recover exposure variations between the frames in the sequence. Although the radiometric response of the camera or film is needed to recover relative radiance values, it need not be known in advance if all you need to do is remove vignetting from your images. Details are given in the ICCV '05 paper.

Below are some results using this technique. With the name of each sequence I note whether I used the constant-exposure or varying-exposure version of the algorithm, and whether the response curve was known or unknown. The first column shows the original images, aligned using AutoStitch and composited without blending (ie. A over B). The next shows the results of Photomontage blending, which doesn't remove vignetting artifacts but does a good job of hiding seams. The third column shows the results of vignette correction without blending. The fourth column shows vignette correction and blending.

Aligned Originals Vignette/Exposure Corrected Photomontage Blending Only Both Vignette/Exposure
Correction and Blending
Allen Center
(Constant Exposure,
Known Response)
Allen Center
(Varying Exposure,
Known Response)
Senore
(Constant Exposure,
Unknown Response)
Green Lake
(Varying Exposure,
Unknown Response)
Apollo 11
(Varying Exposure,
Unknown Response)
Apollo 17
(Varying Exposure,
Unknown Response)

Publications

Dan B Goldman and Jiun-Hung Chen. "Vignette and Exposure Calibration and Compensation." Proceedings of ICCV '05, Beijing, China, October 2005, pp. 899–906. [PDF, BibTeX]

Dan B Goldman and Jiun-Hung Chen. "Vignette and Exposure Calibration and Compensation." Poster. Symposium for Computational Photography and Video '05, Boston, MA, May 2005. [PDF]

Downloads

By request, I have placed my project data (source images, corrected images, and panoramas) online (Gzipped TAR file, 379M).

I am also making the python source code available under the X11 license. It's probably too slow for practical use, but if somebody out there would like to make a more usable version, here's a starter kit! (Gzipped TAR file, 168K).

Links

Acknowledgements

Portions of this work were initiated while Dan Goldman was a research engineer at Industrial Light and Magic, a division of Lucasfilm.
Dan Goldman
dgoldman at cs dot washington dot edu