OCR for Line Drawings?
May 14th, 2009
MIT Press has kindly given me permission to put my first book, Practical Parallel Programming, up on the web. Many of the specifics are out of date, but I think (at least, I hope) much of the discussion is still useful.
One problem, though: nobody has the electronic source for the book’s hundred or more diagrams. I spent a couple of weeks sketching them with a pencil and straight edge in the VU library in the summer of 1994; a graphic artist in Boston then re-drew them using a Mac, but those files are long gone. I could chop a copy of the book and scan what I need as PNGs, but it got me wondering about OCR software for vector images. If anyone can recommend something, I’d welcome a pointer.
Inkscape can generate SVGs from PNGs, maybe with a bit of scripting, the process can be automated.
Thanks for freeing a book!
AutoTrace might be a good first step:
http://autotrace.sourceforge.net/
Vector magic (http://vectormagic.com/home) can convert your images to svg. I don’t think it will actually save any labor, but it will put it into a nicer format that’s easier to edit. They have a handy web interface so you don’t even need to download the software to try it. I’ve used it once before and it did a pretty good job.
Adobe Illustrator does a pretty good job of turning hand sketches and other sorts of scans into vector art. It’s not free but I’m sure the university has discounts for it…
Would bitmap to vector convertors be enough? One I remember is potrace:
http://potrace.sourceforge.net/
If I understand correctly what you want to do, the keyword is “tracing”. Some free tools:
http://potrace.sourceforge.net/
http://autotrace.sourceforge.net/
and probably there are many more. Can’t say whether the quality vs. size tradeoff is acceptable in figures generated by these.
Corel Draw comes with a tool to do this called PowerTRACE.
http://www.corel.com/servlet/Satellite/ca/en/Product/1191272117978#trace
I don’t know what the quality is or if there are open source options. But maybe that gives you a starting point.
You are looking for “vector tracing” programs. There are some free ones out there if you hunt around. Vector Magic (originally a stanford project that spun out to a commercial version) lets you do some free traces via a web app so you can see how well it might work. If it turns out to work well then the paid desktop version might be a better choice for bulk tracing.
IIRC Inkscape has a tracer embedded these days. I haven’t used that kind of stuff in a few years (and then I used Corel), but if you’re mostly talking about line drawings with a bit of text, you should be able to get the line drawings fairly easily and then just re-type the text.
HTH
I’ve used Photoshop’s edge detection filters when scanning maps. I’m not sure how useful it is in creating separate line objects.
I think ‘sharpen’ can be used too.
http://www.autotracer.org is another online service for image vectorization. The results are not as good as those of Vector Magic. But therefore it is still free.