Some time ago I discussed with Rob how to best fit splines to polylines.
Well, I've finally got good results:
I tried various techniques, but finally had to go with the state of the art approach, which took me quite a while to implement. For comparison, the next best approach produced the following results:
Now it would become possible to make adjustments to the auto-detected curves rather than re-create them from ground up if they are not perfect. Well, that's after I clean up the code, which I suspect will take even longer than writing it in the first place.
Spline fitting
Moderator: peterZ
Spline fitting
Scan Tailor experimental doesn't output 96 DPI images. It's just what your software shows when DPI information is missing. Usually what you get is input DPI times the resolution enhancement factor.
-
- Posts: 596
- Joined: 06 Jun 2009, 23:57
Re: Spline fitting
The image looks very good. I've started reading the paper, and it talks about creating a geodesic surface. It seems to me that doing this for a 2D image of a 3D distorted book page would involve evaluating the distance between the splines to infer depth. I haven't read the whole paper, but is that what you're doing? Or is there something about the algorithm that generates 3D splines in some other way? The fact that you say these splines will be user-modifiable suggests to me that they're 2D rather than 3D.
If an answer would require summarizing too much of the paper, don't worry about it, I'm just curious.
If an answer would require summarizing too much of the paper, don't worry about it, I'm just curious.
Re: Spline fitting
Wow! Excellent, excellent work. Is this in git yet?
Scan Tailor is really pushing the state-of-the art. Very impressive. I'm sure your code has applications well outside of book scanning, too. Perhaps in a few years, we'll see a ScanTailor library
Scan Tailor is really pushing the state-of-the art. Very impressive. I'm sure your code has applications well outside of book scanning, too. Perhaps in a few years, we'll see a ScanTailor library
Re: Spline fitting
Hello friend tulon very good job will you implement this in beta 09? A hug
Re: Spline fitting
I don't really know what a geodesic surface is.spamsickle wrote:I've started reading the paper, and it talks about creating a geodesic surface.
The paper I linked to describes a technique for fitting a parametric surface (which is a 3D equivalent of a spline) to a model surface, which might be given as a polygonal mesh or as a (differently) parametrized surface. The technique actually works for any dimensions.
Here are a couple more (one, two) papers from the same authors where they apply this technique to splines. These two miss a few important formulas though, compared to the one I linked to in my first post.
Scan Tailor experimental doesn't output 96 DPI images. It's just what your software shows when DPI information is missing. Usually what you get is input DPI times the resolution enhancement factor.
Re: Spline fitting
Not yet. I'll need to clean up the code first. Then I'll need to integrate it into the GUI.ahmad wrote:Is this in git yet?
Maybe in beta9 or maybe in one of the following betas.emmerick wrote:Hello friend tulon very good job will you implement this in beta 09?
Scan Tailor experimental doesn't output 96 DPI images. It's just what your software shows when DPI information is missing. Usually what you get is input DPI times the resolution enhancement factor.
Re: Spline fitting
Tulon wrote:Not yet. I'll need to clean up the code first. Then I'll need to integrate it into the GUI.ahmad wrote:Is this in git yet?
Maybe in beta9 or maybe in one of the following betas.emmerick wrote:Hello friend tulon very good job will you implement this in beta 09?
ok friend I'm waiting anxiously. Again a great job. Thanks
- daniel_reetz
- Posts: 2812
- Joined: 03 Jun 2009, 13:56
- E-book readers owned: Used to have a PRS-500
- Number of books owned: 600
- Country: United States
- Contact:
Re: Spline fitting
Those splines look amazing. Congratulations!
- daniel_reetz
- Posts: 2812
- Joined: 03 Jun 2009, 13:56
- E-book readers owned: Used to have a PRS-500
- Number of books owned: 600
- Country: United States
- Contact:
Re: Spline fitting
Wow, Michael Kass started the "snakes" approach? Kass is one of my heroes. He recently introduced this technique for manipulating images through smoothed local histograms - it totally reworked my understanding of blur, filtering, etc. Since he works for Pixar, I was able to spend some time with him at a company mixer after SIGGRAPH - super nice guy, too.from the paper wrote:The origin of this technique is the seminal paper by Kass et al. [24], where a variational formulation of parametric curves, coined snakes, is presented for detecting contours in images.
Anyway, again, great work on this, Tulon! (And Rob!) It looks like you've really taken dewarping another big step forward, and that's no small feat.