Countless great photos of people's pets have been ruined by glowing eyes caused by flash photography. Pet owners can always try and fix them with an application like Photoshop, but that's expensive, time consuming, difficult to do right, and not always very effective. What if a website or app was developed that could automatically fix these photos, with little to no effort required by the user?

Software that automatically fixes "red eyes" on people in photos has been around for a while, so there should probably be a way to automate this for pets, too. However, the problem seems much more pronounced in animals than with humans, and there is a much wider array of eye types in pets than humans, so it could be a much more difficult challenge.

The ideal solution would be one that's totally automated, where the user simply uploads or browses to the photo they would like fixed, and the software then produces a fixed version of the photo automatically. If user input is required, it should hopefully be limited to simply clicking on the glowing eyes of their pet to identify them for the program, or maybe uploading a couple of additional sample photos of their pet that were taken without flash photography and look normal, so that the software has something to model its corrections with.

One possible path forward would be to take lots of pairs of photos of pets with and without a flash in rapid succession, in order to create a large training set of photos. Machine learning techniques could then be applied to the training set in order to analyze the effects of flash photography on pet eyes, and to determine how glowing pet eyes might be reversed algorithmically.

Prize Goals:

•   Create an application or website that automatically fixes the "glowing pet eyes" in photos that are caused by flash photography.
•   If any additional input is required from the user, it should ideally be limited to clicking on the glowing pet eyes in the photo, or uploading sample photos of their pet that were taken without flash photography.
•   The software would ideally be available to the public for free, as either an ad supported website or as open source software.

All images courtesy of Wikipedia.