Posted by
Kenneth Sloan-2 on
Jan 18, 2016; 1:19pm
URL: http://imagej.273.s1.nabble.com/Lenght-measurement-tp5015403p5015409.html
This is a classic (and ancient) problem. First - will you be automatically segmenting the nuclei? If you are instead doing manual segmentation, the right answer is to have a trained observer draw the axis as well.
If you are automatically segmenting, then you need to also automatically estimate a so-called “medial axis”. In the distant past, this was known as the “Blum Transform”, which was invented for precisely this problem. It’s essentially what you get by skeletonizing.
The problem is that the medial axis is simple (and *almost* what you want), but in practice can be complicated by high frequency features of the segmented perimeter. Someone has already mentioned blurring the image - I might suggest simplifying the perimeter.
I would probably try to bound the nuclei with a simplified polygon (putting a minimum length on each segment, and perhaps other constraints as well - motivated by knowledge of what a nucleus is “supposed to” look like). But, that is far from an already built-in ImageJ function.
My best idea (assuming all nuclei look like the one shown) would be to fit a quadratic to the pixels in the nucleus. Think of this as a cloud of data and find the best fitting quadratic to that cloud of points. Then, find the points along that quadratic which intersect the boundary. Finally, measure the length of that quadratic “line segment”.
You can do something similar with the skeleton - find the longest path subject to a constraint on the curvature, and then extend this (perhaps just extend the first and last segments) to the boundary.
Bottom line:
smooth the segmented nucleus,
skeletonize,
find a candidate piece of the skeleton that is long, without sharp turns
extend this to the boundary on either end
measure the length
--
Kenneth Sloan
[hidden email]
Vision is the art of seeing what is invisible to others.
> On Jan 18, 2016, at 04:19 , Alexander V. Kalachev <
[hidden email]> wrote:
>
> Dear ImageJ users,
>
> for our current project we need to measure length of spermatozoa nucleus. It is curved in shape
> (see example image attached, nucleus outlined with white dashed line) and we have no idea how to
> measure its length. Could anyone provide us with an advice how to measure the length of such curved
> figure?
>
> With kind regards,
> Alexander
>
> --
> ImageJ mailing list:
http://imagej.nih.gov/ij/list.html> <example.png>
--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html