[vtkusers] RGB image based volume data

[David.Pont at scionresearch.com]

I have used vtk and Paraview for some time primarily to visualise polygonal
data but we are now collecting data which is 'volumetric' and I would like
advice on suitable data formats, filters and techniques for visualising it.

We are a forestry research institute and are destructively sampling the
stems of whole pine trees by serially slicing them and imaging the cross
sectional discs. The trees are about 12m in length, sliced at approx, 35mm
intervals yielding in the order of 3-400 disc images per tree (currently
have 8 trees, more to come!). Discs are imaged sitting on a reference frame
with control points to allow perspective correction and to establish scale.
Images are full RGB at 5634 by 3753 pixels although the region of interest
in the frame (the largest disc) occupies more like 1000 pixels square.
Preprocessing is used to apply perspective correction, crop (and
downsample) to produce an image 'stack' that we want to visualise and
process using Paraview and/or vtk.

I have had some success with this data set using Volview to load and
visualise as an image stack, and saving as a vtk dataset which Paraview can
load.
Issues apparent so far:
- The size of the stack can be huge, but ideally we would like to retain
resolution, at least to zoom in on a ROI. This may well require a custom
built app to handle multi-resolution.
- The volume is inherently RGB rather than a 'typical' volume which has a
scalar value at each pixel, yet there seems to be recognition of RGB (by
Volview) but I am failing to understand what can be done with this type of
data.
- Volume filters, eg contouring, presumably operate on scalar values, is it
possible to operate on 3 component data?  Having said that most of the
useful information is in the red channel.

We also have software to 'segment' the images: extracting the disc from the
background and then delineating boundaries of the growth rings on each disc
surface, each boundary is digitised at 700 points around each ring (1/2
degree polar intervals). This is more familiar territory for me and I have
written code to connect these points and create polygonal surfaces
following growth rings from disc to disc (loosely speaking a set of nested
polygonal cones) and to create hexahedral elements for a 'solid' stem. I
used unstructured grid for these.
I have also played around with using the boundaries to create 4 component
images, for example setting alpha to 0 outside the disc boundary to remove
the background from the RGB images.

Our goals are still a bit vague as this type of data is new to us, but at
least:,
- View the RGB image stack: Volview does a nice job of showing the
individual (or grouped) discs, and 'longitudinal' slices in RGB. This kind
of capability is definately useful.
- View the volume. Volview was doing some interesting things here but I was
very confused about what it was doing and what it might be able to do.
- Segment the volume using edge detection / contouring etc - help me here,
what can be done?
- View the polygonal data from the manual segmentation - this I'm
reasonably comfortable with.
- Combine high res images, volume and polygonal data sets in a single
visualisation! Probably using transparency to blend the 3 data sources.

Finally any recommendations on hardware that will be good for all this. My
boss might go to $20 - 30K, $10K would be better. Any suggestions: my naive
thoughts are: a small cluster? too complicated?, a grunty dual core? and
some kind of stereo monitor sounds good.

Sorry for the looong post, but any help with these queries will be greatly
appreciated.

  regards, Dave Pont

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.vtk.org/pipermail/vtkusers/attachments/20090422/5c453850/attachment.htm>