A web comic discovery tool that solves the copyright infringement problem(justthefirstframe.com)
justthefirstframe.com
A web comic discovery tool that solves the copyright infringement problem
http://justthefirstframe.com/about.html
1 comments
This should be pretty trivial with OpenCV etc.
Unfortunately, nothing in computer vision's ever as trivial as you might imagine, or as it should be.
But, but... square detection!
True. But I bet you'd get 80% of the way there with canny + hough.
I ended up spending most of the day on it and it's more like 20% with that combo.
Fortunately, I found a better way that does get me to 80% or so.
Fortunately, I found a better way that does get me to 80% or so.
No kidding?
What's working?
What's working?
Full writeup coming soon! (I'll comment here with a link, in case it doesn't make front-page.)
Getting the lines is not the hard part, but rather, reasoning about the regions enclosed by them to find the panels. I'm not totally satisfied with my current solution (it's a bit hacky), so I'll try to tweak it a bit, but in any case, results will be posted soon.
Getting the lines is not the hard part, but rather, reasoning about the regions enclosed by them to find the panels. I'm not totally satisfied with my current solution (it's a bit hacky), so I'll try to tweak it a bit, but in any case, results will be posted soon.
so you're trying to solve the "upper corner square" of an item, rather than the "approximately first third" of a linear one?
also, i wonder what his UI works like? i bet you can do this in two clicks of a human per comic.
also, i wonder what his UI works like? i bet you can do this in two clicks of a human per comic.
Thanks joshu and swah! I'm actually the creator of the site and am doing this completely manually. I've explored here and there on how I might automate it, though it all looks a bit daunting. I hadn't heard of OpenCV, and I will definitely explore this further. Thank you so much!
OpenCV is just an API, might be a lot of work.
I think you want to do a Hough transform and then look for long vertical lines.
But I'm no expert.
I think you want to do a Hough transform and then look for long vertical lines.
But I'm no expert.
I bet this is pretty automatable, too.