Zachary Beane wrote: > > > > This is a feature I saw in photoshop. When reducing the colors of an > > > > image, it will try to preserve colors within the active selection > > > > more than it tries to preserve those outside of it. This is one issue... > > > I'd love to be able to select > > > those colors before conversion to hint to the Indexed conversion, > > > "these exact values are important to me." And this is a different one. Now you've made me break out my email client six months ahead of schedule... you'll be sorry. [sven n.] > > Talked to Adam about that issue on #gimp lately and he seems to be > > biased towards changing the conversion-algorithm to be intelligent > > enough to do the right thing without user interaction. According > > to the latest ChangeLog entries, he seems to be actually working on > > this. > > how can the conversion algorithm know what the right thing > is? If I'm going from many colors (one of which is a particular shade > of mauve that matches my web page) to, say, 11 colors, how can it know > that my mauve is one of the colors to preserve unmodified? It can't, and that's why it's a different issue. I assume that Sven was responding to the former of your requests; I did not discuss the latter. I am not enamored with the idea of manually hinting at areas of image to be biased in the colour histogram. There are a number of reasons for this. The foremost is that it's simply a cop-out. The only reason that someone would want to do this would be if they thought that GIMP wasn't doing a good job *without* hinting. I'd rather address the root cause there; there's still lots of room for improvement without resorting to a crutch that naive users might cheerfully wield to their own detriment. Abusing the metaphor, we'll fit the crutch when the leg is mangled beyond healing. As to the other point, I've always actually been dead keen on the idea of 'fixing' colours that the user wants inviolate; the algorithm can't guess the user's own nefarious reasons for wanting a particular colour fixed in this way. This is actually somewhat harder to do the *right* way than the former biasing issue, but it's quite doable. I'll hopefully get around to it one of these years. (It requires algorithm plumbing to do right, though, if someone else wants to try -- no 'quantize to N-M colours and then slot in our M fixed colours afterwards' hacks.) FYI my local tree already has the changes in for quantization, mapping and dithering in CIE L*a*b* colourspace (assuming sRGB source since we don't have any sort of colour management), but also bugs, octopi and other creeping horrors. --Adam -- Adam D. Moss . ,,^^ adam@xxxxxxxx http://www.foxbox.org/ co:3 i l1x0r u! 5luRp!
