Just a quick note to tell that I've updated my project proposal on the wiki: http://wiki.openstreetmap.org/wiki/GSoC_Student_Applications_2009#Draft_application:_Automatic_Street-Sign_Detection_and_Reading
Since SIFT probably isn't going to work, we need something else. The color-histogram-matching might work. While going through my computer-vision book searching for ideas I found a number of other ideas (structural pattern recognition, texture-based features, more segmentation-oriented methods). More details and references in the proposal. I mean, come on, one of these methods must be able to detect street signs. :) - Tijs On Sat, Apr 4, 2009 at 3:38 PM, Stefan de Konink <[email protected]> wrote: > Matt Amos wrote: > >> But please keep in mind that SIFT will *not* work for the typical >>>> streetsign because SIFT looks for features, which are typically not >>>> present in the 'easy to understand' signs (because they use colours). >>>> >>> >> wouldn't SIFT (or variants thereof) work really well against a >> database of exemplars, i.e: pictures of existing street signs, >> isolated letters from the font used for street signage? >> > > http://sift.openstreetphoto.org/templates/C1.png > > This thing is totally 'unsiftable'. > > i'm happy to advise on this project, but i have to state up-front that >> i don't have the experience that stefan has with SIFT, etc... >> > > Just add your name to the list :) > > > Stefan > >
_______________________________________________ dev mailing list [email protected] http://lists.openstreetmap.org/listinfo/dev
