A web comic discovery tool that solves the copyright infringement problem

swah wrote: post #3

joshu wrote: post #2

This is totally clever. I bet this is pretty automatable, too.

This should be pretty trivial with OpenCV etc.

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!

Re: A web comic discovery tool that solves the copyright infringement problem

#5

Post by joshu » Sat, Mar 31, 2012, 12:04 AM UTC

hkuo wrote: post #4

swah wrote: post #3

Earlier quoted context omitted.

This should be pretty trivial with OpenCV etc.

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.

Re: A web comic discovery tool that solves the copyright infringement problem

#6

Post by apu » Sat, Mar 31, 2012, 8:00 PM UTC

swah wrote: post #3

joshu wrote: post #2

This is totally clever. I bet this is pretty automatable, too.

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.

Re: A web comic discovery tool that solves the copyright infringement problem

#7

Post by joshu » Sun, Apr 01, 2012, 8:50 AM UTC

apu wrote: post #6

swah wrote: post #3

Earlier quoted context omitted.

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.

True. But I bet you'd get 80% of the way there with canny + hough.

Re: A web comic discovery tool that solves the copyright infringement problem

#8

Post by apu » Sun, Apr 01, 2012, 11:03 AM UTC

joshu wrote: post #7

apu wrote: post #6

Earlier quoted context omitted.

Unfortunately, nothing in computer vision's ever as trivial as you might imagine, or as it should be.

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.

Re: A web comic discovery tool that solves the copyright infringement problem

#9

Post by joshu » Mon, Apr 02, 2012, 1:15 AM UTC

apu wrote: post #8

joshu wrote: post #7

Earlier quoted context omitted.

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.

No kidding?

What's working?

Re: A web comic discovery tool that solves the copyright infringement problem

#10

Post by apu » Mon, Apr 02, 2012, 1:31 AM UTC

joshu wrote: post #9

apu wrote: post #8

Earlier quoted context omitted.

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.

No kidding? 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.