[Insight-users] registration for tiled volume

[Luis Ibanez]

Hi Rocky,

Thanks for letting us know that the procedure is working for you.

Given that you are setting the origin and spacing correctly, you
could probably skip the careful definition of the FixedImageRegion.
However, you will save *a lot* of computation time by setting this
region because the Metric will be computed using only the pixels
inside that region. If you let the image region open, the metric
will be computed using the entire image, and most of the time
your code will be just testing whether a pixel maps inside of the
moving image.

One way to actually stitch the images together is to create an image
large enough for containing the group of images that you want see
together in a single file, and then use the itkResampleImageFilter,
in order to actually move the intensities to that destination image.

The blending policy is arguable. It is common to do it as a linear
blending where the weights are related to how close the pixel is
from the non-overlapping border of each image. You may have to
experiment a bit on this front, since it depends a lot on how similar
the two overlapping image are (e.g. hopping that there are no
optical distortions or illumination artifacts. Modern microscopes are
quite good regarding aberration elimination, so you probably will not
have any issues of that sort.

We will be very happy to support you in creating a higher level
class for performing stitching. There is a growing number of people
using ITK for microscopy applications and they will certainly be
interested in such new classes.


   Please let us know if you have any questions.


     Thanks



      Luis


-------------------
Rocky Rhodes wrote:

> Thanks.  I've already begun experimenting with just this procedure.  I
> started out by having each block use the same image coordinate space.  I
> selected the overlap region as the FixedImageRegion and set initial
> parameters for the search to position the blocks approximately correct.
> This seemed to give me reasonable results.
> 
> Then I tried using the MetaImage format to carry the consistent image
> coordinates for the blocks.  With this information added, the registration
> class does its job without me setting the FixedImageRegion parameter.
> Should this still be set or is the registration framework smart enough to
> only use the spatially overlapping region for its metric calculation?
> 
> The individual files I have from this device are smaller than I'd like.
> What would be the most straightforward way to stitch them together into
> larger blocks, once I've determined proper transformation information with
> the registration classes?  And then I wonder if there would be any other
> issues.  For instance, after stitching together the first two blocks, the
> axis-aligned bounding box enclosing the two blocks may now contain some
> voxels that are outside the coverage of either of the original blocks.  When
> blending this new block with a third block, I need to make sure that only
> the "full" voxels are blended with the third block.
> 
> Higher level classes for microscopy applications seem like a good idea to
> me.  After I get a little experience with this first pass, maybe I can help
> out with it.
> 
> 	Rocky
> 
> 
>>-----Original Message-----
>>From: Luis Ibanez [mailto:luis.ibanez at kitware.com]
>>Sent: Friday, September 24, 2004 10:50 AM
>>To: Rocky Rhodes
>>Cc: 'insight-users at itk.org'
>>Subject: Re: [Insight-users] registration for tiled volume
>>
>>
>>
>>Hi Rocky,
>>
>>There is not an specific functionality in ITK for
>>registering tiles of images. However, several users
>>have already attempted this for microscopy and satelite
>>images applications.
>>
>>What you can do is to register the blocks by pairs
>>using the registration framework described in the
>>ItkSoftwareGuide.pdf and in the tutorials,  then
>>iterate over the block pair combinations until
>>convergence.
>>
>>It probably makes sense to develop a higher level
>>class that will do this for N1 x N2 x  ...Nn images,
>>I would guess that people in microscopy will be happy
>>to have such a class available.
>>
>>Note that while you do the registration by pairs you
>>may want to set the "FixedImageRegion" of the Metric
>>to be just the region of overlap between one cube and
>>the next.
>>
>>It is extremely important that you make sure that the
>>image cubes have each one consistent image origin
>>coordinates and consistent pixel spacing values.
>>
>>
>>  Please let us know if you have further questions.
>>
>>
>>    Thanks
>>
>>
>>       Luis
>>
>>
>>---------------------
>>Rocky Rhodes wrote:
>>
>>
>>>I have a large volume data set that is composed of a number 
>>
>>of cubes that
>>
>>>overlap a bit.  I need to combine these pieces into a 
>>
>>whole.  Are the itk
>>
>>>registration classes appropriate for this?  I scanned the 
>>
>>registration
>>
>>>chapter in the book and didn't see anything that directly supported
>>>overlapping tiles.
>>>
>>>Thanks for the help.
>>>
>>>	Rocky
>>>
>>>_______________________________________________
>>>Insight-users mailing list
>>>Insight-users at itk.org
>>>http://www.itk.org/mailman/listinfo/insight-users
>>>
>>
>>
>>
>>
>>
> 
>