"extra" junctions in analyzeskeleton

[thejasvi]

Hi,
I'm analysing network images using analyzeskeleton, and keep finding that
the number of reported junctions tends to be more than those that you can
'see'.
Ignacio tells me this is because of the 'slab' regions created when two
thick lines crossover- the plugin treats this as a separating region.

I've also tried dilating the binary image to see if things get better, but
have had no such luck.

eg. the first link has an image with 7 junctions that I 'see' and count
(https://drive.google.com/open?id=0B-rB2njADvZQVGItb1pzb0pEN0k&authuser=0)

     the second link has the skeletonised output which gives 8 junctions in
the table
(https://drive.google.com/open?id=0B-rB2njADvZQRVRWcm1qb3g5QlU&authuser=0) -
this is because of the

Any ideas on overcoming this please ?

thanks,

Thejasvi



--
View this message in context: http://imagej.1557.x6.nabble.com/extra-junctions-in-analyzeskeleton-tp5010607.html
Sent from the ImageJ mailing list archive at Nabble.com.

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Gabriel Landini]

On Monday 24 Nov 2014 19:23:29 Thejasvi wrote:
> I'm analysing network images using analyzeskeleton, and keep finding that
> the number of reported junctions tends to be more than those that you can
> 'see'.
[...]

You are using the wrong approach to the question you want to answer.

The skeleton analysis program works fine, but in your image the "lines", when
skeletonised do not generate straight lines. That is the reason you get extra
joints in the skeleton intersections.

One possible way round is to try to extract straight lines with something like
the Hough Transform and once you find the lines, compute their parametric
description (ie. y=ax+b) and then find the intersections with other lines via
computational geometry:
http://www.mathopenref.com/coordintersection.html
You then cycle through all lines and find the intersection with all other
lines. You will need to check if the intersections fall within the image frame
too.
I am not aware of code that does all this, but at least you get the idea about
how to proceed.

Regards

Gabriel

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[thejasvi]

Dear Gabriel,
Thanks so much for the pointers!
This approach is quite intuitive, I'll try and see if it works.


thanks again,

Thejasvi