Hi all,
I am trying out TrakEM2 to see how useful it is for aligning and analyzing
serial sections of plant material. So far, I am quite impressed by its
capabilities! I have encountered a few issues and hope someone can offer
some tips:
I find the the alignment algorithms sometimes have problems with making a
good alignment. This seems mostly due to some damaged or missing sections.
I assume it fails to select "correct" features for the alignment. Since I
have to do manual segmentation of the image anyway, why not simply use this
human-input for alignment? However, this seems impossible currently:
AFAICT alignment using landmarks (is there a way to undo a misplaced
landmark other then by editing xml of saved landmarks?) only allows for
manual alignment, not the nice elastic algorithm. It also forces me to
define points instead of outlining the cells.
To overcome this I decided to define cells and organelles first using
arealist, and then try to do the alignment on the segmented arealist. It
seems that to make this work, I first need to export the arealist as
labels, then reimport this as an image. I had some success doing this, but
I am not sure the parameters are tuned to work with images containing only
the segmented surfaces (homogeneous colors) of interests. Perhaps someone
can comment on whether such method can work).
It seems in the past (manual and examples) there was an "alignment using
profiles" option. This sounds like it might to what I want (except using
profiles instead of arealists, but the difference between the two is not
clearly described in the manual), but it is no longer available in the
versions on github?
I could also try to spend more time on aligning the original image. In this
case, it would help if arealists can be linked to the image so that if I
manage to redo the alignment using the original image data, it will apply
the same non-linear transformations to my arealist? I see no reason why
this should not be possible, but have no idea how to script it using the
jython terminal. (this seems a useful feature anyway since the segmentation
is by far the most time-intensive, and any change to the original data
seems to require you segment the full stack again).
Cheers,
Danny
--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html
Locked
