[Insight-users] Registration?

[Linglan Edwards]

Luis,

	To answer your questions:
	
>Date: Mon, 04 Nov 2002 11:58:54 -0500
>From: Luis Ibanez <luis.ibanez@kitware.com>
>User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:0.9.8) 
Gecko/20020204
>X-Accept-Language: en-us
>MIME-Version: 1.0
>To: Linglan Edwards <linglan@lhc.nlm.nih.gov>
>CC: insight-users@public.kitware.com
>Subject: Re: [Insight-users] Registration?
>Content-Transfer-Encoding: 7bit
>X-Scanned-By: MIMEDefang 2.21 (www . roaringpenguin . com / mimedefang)
>
>
>Hi Linglan,
>
>Some question about your registration:
>
>1) Are you performing this registration entirely in 2D ?

Yes.

>
>2) Are you registering the cryogenic data to itself ?
>    that is, RGB to RGB ?

	Yes. If you could help, that'd be great.
>
>    ITK current metrics do not manage RGB pixels directly.
>    It will be quite easy, however, to create a new metric
>    based on existing ones.  I'll be glad to help you create
>    a new metric supporting RGB data.
>
>3) How large do you estimate the missalignment
>    could be ?

     It varies. I'd say anywhere from 10 pixels to 100 pixels.
>
>    If the missalignemen is large, you may want to use
>    the multiresolution framework. However this will
>    add memory requirements to the problem since a
>    pyramid of images is created. If the missalignment
>    is small, a direct registration could be sufficient.
>
>    Each slice in RGB will make about 265 Mbytes...
>    In the registration process, ITK metrics are the
>    only components that directly interact with the
>    images. The metrics are not streamed at this point,
>    although some of them they could be.
>
>    If you have about 1 Gigabyte of memory, you could be
>    able to run the single level registration 2D for these
>    slices.

	I do have 1 GB memory (on Sun Ultra 60).
>
>4)  We could easily modify the registration framework
>     to evaluate the metrics is a set of selected regions.
>     Rigth now, you provide a specific rectangular (ND)
>     region on the fixed image.  This could be replaced
>     with a set of regions, in order to require only one
>     full slice to be on memory for the moving image,
>     while the fixed image will be loaded as a set of
>     smaller regions of interest.
>
>5) Another option is to select a population of points
>    from the moving image and perform PointSetToImage
>    registartion. In this way the points could be more
>    sparsely distributed on the moving image than a
>    set of regions proposed in (4).
>
	Option 4 sounds like a better solution in our case...
>
>Please let us know what options do you consider
>reasonable for your registration problem,
>
>
>    Thanks
>
>
>       Luis
>
>
>==========================================================
>
>Linglan Edwards wrote:
>> I'm looking for a fast way to aligned some of the Visible Human 70mm scanned 
>> data. I think all I need are some rigid transformations. What ITK modules 
shall 
>> I use to do that? Also, the images are very large (each slice is 
>> 4096x2700x24bits), will ITK be able to handle it?
>> 
>> Thanks,
>> 
>> Linglan
>> 
>> 
>> _______________________________________________
>> Insight-users mailing list
>> Insight-users@public.kitware.com
>> http://public.kitware.com/mailman/listinfo/insight-users
>> 
>> 
>
>
>
>
>
>

Thanks,

Linglan