[SciPy-user] Advanced Image Processing with Python

Zachary Pincus zachary.pincus@yale....
Tue Jun 24 09:11:16 CDT 2008

> Hi Zach,
> Among all Python image processing libraries, ITK appears to be the  
> most promising (thanks to WrapITK - as you know, Zach, of course: http://voxel.jouy.inra.fr/darcs/contrib-itk/WrapITK/WrapITK_-_Enhanced_languages_support_for_the_Insight_Toolkit.pdf) 
> . But for Windows users (and I am one, unfortunately), ITK seems  
> difficult to build compared to VTK for instance (and it's a very  
> long process - 3 hours on my machine which is quite powerful). I've  
> already tried and failed with MinGW and VSC2003.
> That is really weird for such an interesting library to be almost  
> closed to Windows users: there is no binaries available for Python  
> 2.5 (actually there is none for VTK either but at least it's easy to  
> build). And when I see what it can do (at least on paper), it  
> surprises me that no one tried to build it on Windows before (and  
> shared the binaries with the community), except for Python 2.4 (http://cpbotha.net/2007/08/02/python-enabled-vtk-51-and-itk-32-windows-binaries/ 
> ). Is there any reason for that? (is this library not so good as it  
> seems?)
> Anyway, if I succeed, I will be glad to share the result with the  
> community.

Hi Pierre,

I'd advise asking for help on the ITK list -- they take quite  
seriously their commitment to cross-platform build-ability (to the  
extent of having built that CMake tool precisely to that end).

As for build time, that's another issue. Basically, ITK is (to a  
frightening degree) driven by C++ template programming, which really  
strains compilers. You can speed the build time for WrapITK (when you  
get it working) by turning off the template instantiations for all but  
a few pixel types (say, uint8, uint16, and float32) and image  
dimensions (only 2D, say). Build time will be nearly linear in the  
product of pixel types and image dimension...

Now, it's been a while since I was involved with WrapITK, but last I  
checked, the ITK people were pretty enthusiastic about getting it  
integrated into ITK and fully supported. Hopefully this is still the  
case! (Though coaxing the CMake build system into performing the magic  
required to run everything was non-trivial, due to limitations in the  
CMake language, so perhaps it has been difficult to maintain.)

Good luck,


More information about the SciPy-user mailing list