Gregory Williams wrote: > I've just taken a look at the Canterbury tile. It looks good, although > I'd observe that there is a noticeable horizontal shift in places. Look > at Stone Street, for example (That's the B2068 Roman Road south of > Canterbury for non-locals on the list.). > The accuracy is limited by several factors but the most significant are OS survey techniques and the division of the map into tiles. I notice the tiles around stone street (613,146) is about 5 pixels out of alignment horizontally which is causing the discrepancy in this case..
It may be worth reassembling the NPE map (from tiles or backup) and retiling, it but it would be a big job! (The potlatch blog mentioned anchor points and improving alignments...) Richard Fairhurst wrote: > Could you tell me how you're doing it? > I am doing a piecewise linear affine transform on each NPE tile to go from GBOS to EPSG:4326 for use in JOSM. The linear transform is a least squares fit ... well actually I divide each tile into two triangles and transform those. The whole thing is implemented in PHP and Imagemagick. It takes about 60 seconds to render one 0.1 by 0.1 tile - pretty slow! The GD library is much faster but it does not seem to have an affine transform function. I did see you are making NPE available on Potlatch. What projection is this in? I am guessing spherical Mercator. > As posted previously I'm currently reprojecting the 300dpi originals > into spherical Mercator, for use in anything which uses the same tile > system as OSM. It would be a bit silly if we were both doing the same > thing. > I am using the jpeg tiles for reprojecting - which is lossy and not ideal... Is the 300dpi data available on the dev server or perhaps elsewhere? Tim _______________________________________________ Talk-GB mailing list [email protected] http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/talk-gb
