speed of iradon transform

Stefan van der Walt stefanv at berkeley.edu
Mon Apr 20 13:48:45 EDT 2015


Hi Matt

On 2015-04-20 04:59:06, Matthew Newville <matt.newville at gmail.com> 
wrote:
> Some preliminary timing results:

Thank you very much for your detailed investigation!

>    numpy      skimage    workspace?   Best, Worst of 5 (s) 
>  |---------+-----------+------------+---------------------| 
>    1.9.2      master      N/A            6.085  6.132 master 
>    master      N/A            1.784, 1.813 master     PR1474 
>    No             1.788, 1.809 master     PR1474      Yes 
>    1.103, 1.160 
>  |---------+-----------+------------+---------------------|

It looks like one gets about 1.5x for using the workspace.  I am 
always careful about added code complexity for a relatively small 
gain, but in this case your refactoring *improves* legibility of 
the code--so +1 from me.

> Oddly, iradon() is now faster than radon() on this machine/image 
> size  (radon() takes about 3.5 sec).   I don't understand why 
> that would be.   Anybody understand why radon() is so slow?

I'm afraid we don't implement an optimized version of the forward 
radon transform, such as

Brady, "A Fast Discrete Approximation Algorithm for the Radon 
Transform", SIAM J. Comput., 27(1), 2006

A fast digital Radon transform--An efficient means for evaluating 
the Hough transform WA Götz, HJ Druckmüller - Pattern Recognition, 
1996

This overview from William Press of Numerical Recipes fame is 
helpful:

http://www.pnas.org/content/103/51/19249.full

Regards
Stéfan



More information about the scikit-image mailing list