[Insight-developers] Notes on Insight Telcon 4 May 2001

[Paul Hughett]

               Notes on Insight Telcon 4 May 2001


CMake has been having some problems with the tcl scripts and configure
over the past week, but they should be mostly solved now.

The automatic testing has also been having some problems, apparently
due to the fact that the scripts to automatically install the new
copyright text have not been working correctly for tcl scripts.  This
should be fixed soon.

The possibility of distinguishing plain and continuous image functions
was discussed; the primary distinction is that continuous image
functions can be evaluated at any point, whereas plain ones can be
evaluatated only a integer index coordinates.  A second distinction is
whether image functions should take arguments in index coordinates or
physical coordinates.  Lydia Ng will write and commit a working version
for evaluation and possible revision.

There seems to be some overlap of function here and there.  Paul
Hughett and George Stetten are both working on image resampling;
George Stetten and Lydia Ng are both working on continuous image
functions.

Paul Hughett's resampling class should be templated over the
interpolation function and the type of coordinate transform, even
though most image filters are templated over only the input and output
image types.

Using CMake to do projects using ITK but not part of it remains somewhat
problematic.  Bill Hoffman is the point of contact when problems occur.

Someone--I didn't catch the name--is working on improved data
accessors for images.

The Viola-Wells mutual information registration algorithm is being
experimented with; some problems and quirks are surfacing.  It is
important to set the center of rotation; rotating around an off-stage
origin looks a lot like translation and tends to confuse the
optimizer.  Smoothing the images before registration can be very
helpful, and a multi-scale algorithm is being considered.  Using about
50-80 sample points to estimate the intensity PDF seems to work
reasonably well; using more tends to cost too much time.  Choosing the
right variance for the Parzen window seems to be difficult in
practice.  Adding an ROI to limit the sample points to the head or
brain would be a useful improvement.

There does not seem to be a RawImageReader anywhere in the Insight
Toolkit, although there is a RawImageWriter.

The level of compilation warnings to be used can be controlled by setting
CFLAGS (or perhaps CPPFLAGS) before running configure.

Using ConstPointers for the inputs to image filters seems to be a good
idea, but will require that various image metadata used to control the
pipeline be made mutable.

The ITK_EXPORT macro should be used to qualify all classes that might
be included in a shared library and exported to the outside world, but
not classes used only internally.  It is needed to make Windows DLLs
work properly.

These telcon notes should be send to the mailing list but not saved in
the repository.

There is a search engine available in Doxygen which we should look
into.  Doxygen also provides a variety of useful keywords which we
could likely take better advantage of, and support extra (non-source)
files which would be useful for overviewing a set of related classes
or discussing theory.

The bug tracker can also be used for admin issues, feature requests,
and open issues that should not be forgotten.  [Ed note: also
capability limits and performance issues that aren't critical now but
should eventually be fixed.]


Paul Hughett