Dirk Stöcker <[EMAIL PROTECTED]> writes: > Hello, > > when I look at some tiles, then they use totally wrong colors. One example > is http://tah.openstreetmap.org/Tiles/tile/12/2148/1418.png > > Now one could say this is an effect of the color reduction, but when I > look into the image, then I find multiple color entries with the same value, > so > there must be an error somwhere in the processing chain which should be > found and fixed.
You are probably right and if it is a bug it is probably in pngnq. I have plans and a partial implementation to use a color quantization algorithm that ensures that all colors that are defined in the osmarender styles and symbols remain unchanged. This will work as follows: - at startup of the client analyze all style files and symbols and store a color palette for each layer in memory - use the appropriate color palette as base palette for each tile - eliminate from that palette all colors that are not present on that tile - quantize the rest of colors of that tile that are not already in the palette down to the number of colors needed to fill up the palette to 256 colors - transform the tile using the new palette This should prevent the deviation of all colors we really care about. The only exception I can think of are those colors that result from the use of less than 100% opaque colors in the styles that can be virtually any color depending on the color underneath. The accuracy of all other colors which should only result from anti-aliasing is not that important. I am 3/4 there with my implementation. It might take me a week or two to get a first working version to be tested in _unstable. Matthias _______________________________________________ Tilesathome mailing list [email protected] http://lists.openstreetmap.org/listinfo/tilesathome
