suggest algorithm: gradient-fill a polygon
in [General Programming]
|
From: makc on 3 Mar 2010 02:08 On Mar 3, 4:14 am, Nicolas Bonneel <nbonn...(a)cs.ubc.ca> wrote: > - Consider your problem as a minimization problem, where you want to > minimize the gradient in unknown location. This is basically a Poisson > problem, and is solved with any Poisson solver, passing the known points > as dirichlet boundary conditions. You can use relaxations techniques > which have shown to work well on the gpu That's another interesting idea, do you think repeated drawing black edges/white point and averaging 3x3 pixels would do the trick, too?
From: Kaba on 3 Mar 2010 03:14 makc wrote: > On Mar 3, 12:51 am, Kaba <n...(a)here.com> wrote: > > "Diffusion Curves: A Vector Representation for Smooth-Shaded Images" > > > > http://artis.imag.fr/Publications/2008/OBWBTS08/ > > > > Look at the video, its nice:) Just some food for thought, maybe you get > > some ideas from that. I haven't looked into the paper too detailedly. > > > > are you sure the URL is right? it refuses to open here. The URL worked yesterday, but today I can't access it either. -- http://kaba.hilvi.org
From: makc on 3 Mar 2010 04:39 On Mar 3, 9:08 am, makc <makc.the.gr...(a)gmail.com> wrote: > That's another interesting idea, do you think repeated drawing black > edges/white point and averaging 3x3 pixels would do the trick, too? now when I thought about this more, I could see this could also work for my original idea with delaunay mesh to color vertices by repeatedly averaging connected ones. this has less data to operate on, but on my particular platform (flash) there are built-in means to average pixels, which could still be faster... scheduling benchmark :)
From: Kaba on 3 Mar 2010 10:37 Kaba wrote: > > are you sure the URL is right? it refuses to open here. > > The URL worked yesterday, but today I can't access it either. It seems to work again: http://artis.imag.fr/Publications/2008/OBWBTS08/ -- http://kaba.hilvi.org
From: Nicolas Bonneel on 3 Mar 2010 16:13
makc wrote: > On Mar 3, 4:14 am, Nicolas Bonneel <nbonn...(a)cs.ubc.ca> wrote: >> - Consider your problem as a minimization problem, where you want to >> minimize the gradient in unknown location. This is basically a Poisson >> problem, and is solved with any Poisson solver, passing the known points >> as dirichlet boundary conditions. You can use relaxations techniques >> which have shown to work well on the gpu > > That's another interesting idea, do you think repeated drawing black > edges/white point and averaging 3x3 pixels would do the trick, too? This is exactly the operations to do which converge to the Laplacian(u)=0 solution. These are relaxation steps. They have shown to converge... but very very slowly. If you're not concerned with the speed or the accuracy, this is the way to go. If you're concerned with either speed or accuracy, then you have to switch to multigrid methods for example which compute these relaxation steps at different mipmap levels. Or using TAUCS which will solve exatly your problem (no accuracy issues but it can be slow), but you have to build your finite difference matrix. Best, -- Nicolas Bonneel http://cs.ubc.ca/~nbonneel/ |