Few comments to my previous post. I was thinking about how is it possible that cartoonizer if working over layer with previously reduced colors, is sometimes not able to pick right colors. (and all of them)
It seems that my algorithm is good for "normal" images where RGB values are distibuted continuously. Then method of small iterations works well. But if input colors are discrete my method may fail.
So in regard to my workaround I adviced to increase color weight, but it do not guarantee success. But the effect of smoothing sharp angless can be achieved anyway.
thank you for your review, I appreciate your time you spent on this and ideas. I have no login on gimpchat (I had some problems to register there some time ago and gave up...) so I will react here.
First issue I want to talk about is smoothness of outlines. Bit of theory. Plugin have two cleaning techniques. First one is single pixel cleaning and second one (in order as executed during processing) is removal of spots.
I believe the first one (with maybe small modification) can smooth the lines (borders of color areas). To do this this single pixel cleaning would be executed after spot removal. But as a workarond try this:
- Run cartoonizer and export image to new layer.
- Quit cartoonizer, switch to newly created layer and run cartoonizer on this layer.
- With default settings Cartoonizer will change change the image (a bit surprising but it is how it is), but change 'color weight' to maximum. Now calculated image in preview should match the original in layer...
- In single pixel cleaning section change treshold to 5 and iteration to 10 or so.
Is what you got similar to what you needs? This is workaround but I may consider redesigning plugin - technically it would be trivial...
Second - gradient transition between colors - are you talking about simple blur? If it is the case I see no reason to implement it as a part of this plugin, as it is really simple to use blur afterwards. If not, be more specific or provide some examples if possible.
This is all for now, I am interested in your feedback.
(I presume you are using 0.3.x version)