I’m working on image interpretation for the city engine to build rounded, curved roads and sidewalks starting from a non-vectorial image.
it’s not finished yet, but it’s on the way :
bottom-right is the source pixel image and top-left is the result as polys. (roads are only a plane here.).
there’s 2 (main) steps here. the first one interpret each borders as a vector or coords lists, the second one smooth the angles between the discovered vectors.
the difficulty is to find the right vectors, to obtain the same shape :-/
I open this thread apart from sce as maybe it could be useful for other purposes, to rescale textures or make some FX starting from pictures, quickly draw large maps in the uv editor, maybe create a polygon function libraries etc…
the source here is B&W but the process wouldn’t be very different for a color image.
I also post apart because it has perhaps been already done.
the fractureMe script does something similar but the concept is different as in this script the black lines of the source image have no width, where as I need different widthS I can’t ‘resize’ the same way around each polys as there’s different road widths. [maybe someone will understand what I’m trying to say…]. and for sure the goal is not to make a svg clone 
(speaking about svg, any contributions would be great to add this format as input for streets in the sce.)
so if you see any other applications for this image parser please c&c