[ImageJ-devel] Labeling / segmentation proposal for imglib
Lee Kamentsky
leek at broadinstitute.org
Wed Nov 17 08:14:32 CST 2010
Hi Albert,
I could see how using ShapeLists could work well for us for most cases
seeing that we mostly use 2-d. It looks to me like you need one shape
per z-plane in 3-d? I am thinking that there is still some utility in
having something that represents the segmentation or labeling as a
whole. In that scheme, the shapes would be views into the underlying
representation of the labeling. Finally, I think there is some utility
to a localizable cursor that doesn't use shapes as an intermediate
because that sort of cursor would only visit labeled pixels whereas the
shape cursor, at best, visits all pixels in the shape's bounding box
(and for naive uses, might visit all pixels in the image). Some
labelings, such as those on neurons, are sparse within their bounding
boxes and, it seems to me, would require many calls to
java.awt.Shape.contains(x,y).
--Lee
On 11/16/2010 5:02 PM, Albert Cardona wrote:
> 2010/11/16 Lee Kamentsky<leek at broadinstitute.org>:
>> Now I do ;-)
>>
>> I think the ShapeImage container is not appropriate since it's based on
>> java.awt.Shape - I think that segmentation results are too general to be
>> represented by shapes, the ShapeListCache doesn't scale to hundreds of
>> labels per image,
>
> The cache is per cursor, and simply stores the last N desired visited locations.
>
>
>> and the typical algorithms used on these sorts of results
>> might not be efficient.
>
> Have you tried it? I have. Currently, giant (>5000) areas are rendered
> in 3D using marching cubes and it works great.
>
>
>> Also, using java.awt on the Mac will cause
>> difficulties for CellProfiler and others because of some unfortunate
>> implementation decisions in awt and Mac OS beyond our control.
>
> java.awt.Shape is an interface. That shouldn't cause any problems.
> java.awt.geom.Area, java.awt.Rectangle, and others, are in the
> java.awt package but have nothing to do with the GUI awt classes. They
> can be used cleanly.
>
>
> You may want to give it a second look and run some tests.
>
> Albert
>
More information about the ImageJ-devel
mailing list