Quote:
Originally Posted by Retrofan
Doesn't look very good. You can see in your own photo that it's "made with points". Using it's own palette on Fullpalette then all the letters change colors and you don't see anything.
|
Yes, dither is visible as palette is suboptimal from color quantization point of view but colors need to be shared with system and with picture.
Perhaps it can be improved by correct mapping (not sure how both palettes are populated - have no time to analyze this).
Quote:
Originally Posted by Retrofan
Edit: Thinking now, your Riemersma conversion can be the best of all. This time I've made a photo with Imp Pant, then pasting to Paint, then selecting and uploading:
|
Riemersma is similar method to Floyd Steinberg or comparable error spread/ distribution method - simply i like it more than Floyd Steinberg as it can produce better from subjective point of view dither patterns.
Color quantization method was k-means, but in case where you provide external palette this is not valid - colors should be taken from palette and used to produce lowest possible error in terms of distance between color optimal and color available.
Quote:
Originally Posted by Retrofan
|
Still not optimal i think - optimal is when only fraction of colors is used by picture, dynamic colors (system one) should be not used as they can changed - still have problem - i don't know which part of index are used by system and which one not - best is to not use system index in picture.
Software used by me is gimp+ximagic plugin (you need also pspi plugin), to extract palette and some operation (conversion to Paint Shop Pro format) i use xnview - all above seem to be quite standard on typical PC where Adobe is not available.