I need to run CLAHE on a bunch of images, but there are 3 ways of doing it;
- CLAHE on each RGB channel (RGB)
- CLAHE on L channel (LAB)
- CLAHE on V channel (HSV)
I've found however, that all methods work on some photos, and all methods fails on others (and by fail i mean create something worse than the original photo). I need like 99.9% accuracy, so having a 50/50 gamble whether LAB is going to produce beauty or the beast, is not ideal. So is there a way for me to make a script/algorithm look at a pictures histogram, and determine what method would create the most equalized histogram (aka. The nicest photo)
I imagine this would require an AI that we feed the unprocessed histogram as input, give it the 3 options as outputs, and reward it dependent on how flat the output histogram is.
But is there another way? cause I really don't want another AI on this project. We already have 2 xD (It's for business use, for deployment in the cloud, in case it wasn't clear yet, so it doesn't have to be super optimized)