Color Cast Experiment Results

Here are some results of a color constancy algorithm I have been working on.

The data set was collected by Brian Funt's group at Simon Frasier University and can be found here.

Notes on image processing:

Before color processing all of the images were clipped at a brightness level of 100 to better simulate real world exposure levels.
After color processing (to facilitate screen based comparison) all of the images (including the original ones) were processed to have the same mean and a gamma correction of 2.2 was applied.

Notes on image labels:

Original Image = uncorrected (but still clipped) image
Gray World Correction = simple gray world such that all three channel means are equal
Bayesian Correction = combination of EM and Bayesian correction
Optimal = best possible diagonal transfomation determined from a calibration patch in the image

The settings I used to generate these images can be found here.

Last modified: Tue Mar 21 11:39:06 EST 2000