[Insight-users] FW: erosion performance for binary images

[Richard Beare]

Here's some more options for you to think about, depending on your application.

As Gaetan pointed out, performance of many of the specialised binary
filters  is content specific, whereas the ones you've tried so far are
general purpose.

First of all, I did some experiments with decompositions of balls that
should still be in the code somewhere. The problem with these was that
it was difficult to get a size that was close to what you were after -
this is due to the decomposition involving quite a lot of lines, each
of which needed to be an odd number of pixels long. However, if you
are just after a big blob operation, then this might work for you.

Another option for erosions, which should be precise, is threshold of
distance transform of your mask. This will also be independent of ball
size, and potentially quite quick if you have a machine with lots of
cores - the signed distance transform in the new multi threaded
implementation of the parabolicMorphology package performs
significantly better than Maurer in my latest tests, and there is
still a lot of tuning possible.

Finally, are you really after a kernel based operation, or do you want
to discard small objects (a common application or erosions/openings)?
If you are really after the latter, then attribute operations are
worth checking out.

On Sat, Jun 6, 2009 at 8:49 AM,  <M.Staring at lumc.nl> wrote:
>
>
>
> -----Oorspronkelijk bericht-----
> Van: Staring, M. (LKEB)
> Verzonden: za 06-06-2009 00:44
> Aan: Dan Mueller
> Onderwerp: RE: [Insight-users] erosion performance for binary images
>
> Hi Dan,
>
> thanks for sharing your results. VHGW indeed looks good here. Next week I
> will try again at my machine and make sure windows is not secretly running
> other programs while I run my test.
>
> Thanks for all the feedback,
>
> Marius
>
>
> -----Oorspronkelijk bericht-----
> Van: Dan Mueller [mailto:dan.muel at gmail.com]
> Verzonden: vr 05-06-2009 21:32
> Aan: Staring, M. (LKEB)
> CC: insight-users at itk.org
> Onderwerp: Re: [Insight-users] erosion performance for binary images
>
> Hi Marius,
>
> Thanks for providing your code. I ran it with the following:
>     ITK 3.12
>     CMake 2.6.2
>     Visual Studio 8.0.50727.867
>     Windows Vista SP1
>     2 x 2 GHz, 3 GB RAM
>     Release build
>
> The results were as follows:
>
> Box radius: 1
>   Elapsed time erosion_binary: 6.558s.
>   Elapsed time erosion VHGW: 6.177s.
>   Elapsed time erosion Anchor: 44.742s.
> Box radius: 2
>   Elapsed time erosion_binary: 7.068s.
>   Elapsed time erosion VHGW: 6.114s.
>   Elapsed time erosion Anchor: 42.208s.
> Box radius: 4
>   Elapsed time erosion_binary: 8.595s.
>   Elapsed time erosion VHGW: 6.199s.
>   Elapsed time erosion Anchor: 44.865s.
> Box radius: 8
>   Elapsed time erosion_binary: 14.707s.
>   Elapsed time erosion VHGW: 6.46s.
>   Elapsed time erosion Anchor: 44.666s.
> Box radius: 16
>   Elapsed time erosion_binary: 38.651s.
>   Elapsed time erosion VHGW: 7.117s.
>   Elapsed time erosion Anchor: 46.104s.
> Box radius: 32
>   Elapsed time erosion_binary: 125.506s.
>   Elapsed time erosion VHGW: 7.656s.
>   Elapsed time erosion Anchor: 49.61s.
>
> These results match with my previous experience with the VHGW filter:
> constant time, very fast. Thanks again Richard and Gaëtan!!
>
> Marius: I hope this helps.
>
> Regards, Dan
>
> 2009/6/5 Bill Lorensen <bill.lorensen at gmail.com>:
>> Marius,
>>
>> On what platform are you running the tests? And did you build it Release?
>>
>> Bill
>
>
>
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
>
>