[Insight-developers] Fast distance transform - how do I get it into ITK?

[Benjamin King]

Hi,


Apologies first: After submitting a paper on a fast and general  
distance transform to the Insight Journal (http://hdl.handle.net/ 
1926/196), I have left the field of medical image processing.  I took  
a look at the promising reviews from time to time but didn't really  
have the resources to adress them properly. So my apologies to all  
reviewers who waited for feedback. I didn't mean to be rude.

This summer, however, I have spent some time to revisit the code.  
Among other things, the new version uses L1 cache more efficiently  
and uses multiple processors when compiled with OpenMP support.

The filter is now 20 to 30 times faster than  
DanielssonDistanceMapImageFilter on an Intel Core2 CPU. Without using  
OpenMP it still achieves a speedup of 12 to 17 times.

I have measured this for image sizes ranging from 136x136x121 to  
512x512x348, so there is no degradation in performance for large  
images any more.

Given the current efforts to support microscopy image analysis with  
ITK4, I see also a possibility that more users might have use for the  
group morphology filter that becomes possible (see section 4.5 in the  
original paper).

I'd really by happy if this code would make it into ITK, but I wonder  
on how to proceed.

Should I submit a new paper to InsightJournal?
Or rather post the new code as a revision to the old one?
Or just post it here to the developers list for further inspection?

I'd be happy if anyone could give me some advice on this.


Cheers,
   Benjamin

PS: I'm now developing software for railway companies, and I'm not  
affiliated with Hannover Medical School any more.