Limitations of Skeleton L

[b.ploetzeneder]

Hi,

I am stumbling upon limitations of the Skeleton function, and IMHO, they are regarding to feature size.

Let's assume I am workign with a big image.. 2048*something.
What I get from the Skeleton L function is this

(I know, Skeleton lines are barely visible. What matters is the "block in the middle".. aka the Skeleton function is overflowing)

This is the same operation at 600*355px image size (which is NOT large IMHO)

Only with an image size of 230x160 or so, I am starting to get a "real result"

I am not really happy about this behaviour and losing A LOT of detail here.

I have two suspicions:
a) Somewhere data is stored in an U8.. 
b) IMAQ Skeleton is most likely based on thinning skeletonization, which in turn is probably based on IMAQ Morphology. IMAQ Morphology starts being unreliable after a magic 100 iterations.. 

Can anyone from NI give more info? I will probably have to re-implement the entire thing.. but would at least like to know where / IF I can trust this algorithm to a certain particle size.

Thanks
B

[Hatef.fouladi]

http://zone.ni.com/reference/en-XX/help/370281AA-01/imaqvision/imaq_skeleton/
did you regard greater border than 1 for image?

[b.ploetzeneder]

Standard border size for IMAQ Images is 3 anyways.

But yes, I tried now (even with border size >>), and it doesn't help.

[Hatef.fouladi]

could you send your original image to test?

[b.ploetzeneder]

Was already attached to initial post.

[Hatef.fouladi]

this is my answer
L shape algorithm for Ni do not regard your shape as a L shape based on number of shape pixels concentration in area 
so this algorithm is based on size and pixel amount for regarding the shape is L shape or not 
and in that size this algorithm do not regard your shape as a L shape 
I reshape a bit your shape for test this 
check bellow image and you will be see that in this shape this algorithm work better
also I think this is not good that this algorithm work based on number of shape pixels and it should work with percent of pixels for regarding  a shape is L shape or not 

[b.ploetzeneder]

Skeletonization is not about shape recognition. It is about extracting topological information (symmetry, longest path etc) - see here https://en.wikipedia.org/wiki/Topological_skeleton. 

And the "L" signifies which structuring element NI chooses for thinning. Or multiple "conditional" thinning.

[Hatef.fouladi]

my means is this that you can see in this image 
did you check L skeleton on image that I send for you 

[b.ploetzeneder · Accepted solution]

Just for anyone actually searching for the topic: I ended up implementing my own Skeleton algorithm. It works, it has no size limitations (beyond U16 distances :)) and is 84x (on my application) as fast than the NI one. 

https://decibel.ni.com/content/blogs/bp/2016/10/04/skeletons-in-ni-vision-or-lets-do-something-nicer