The Battle of the Algorithms: Testing and User Interface

We began this project on a quest for the best color balancing algorithm for a given image. What we discovered is that this is a much trickier problem than first anticipated. Five basic algorithms, and several combinations of the basics, were evaluated to determine which algorithms performed better for some images and failed for others. Although our image sample size was too small to make any broad generalizations, we did see a few trends. For images in which one of the algorithms that adjusted mean and standard deviation performed well, all of the algorithms that performed a variant of these operations did equally well. For images in which Gray World performed well, Scale by max and White World tended to also outperform the other algorithms. This was shown in the case of the White Flowers image. If we had to choose a winner it would be the algorithm that adjusted the mean and standard deviation. Most of the images color balanced with this algorithm were chosen in the subjective comparison as top winners.

Overall the project was really fun to work on and we eventually ended up with a nice Matlab user interface for evaluating different color balancing algorithms. We hope that it will be put to good use in the future.