It was obvious then too. Way back when I was taking the raw grayscale data before debayering and I immediately saw I'd need to treat each color's pixels separately. No background in compression or signal processing or anything, just a tinker's mindset and some experience faffing around in Photoshop/GIMP