[ImageJ-devel] Labeling / segmentation proposal for imglib

[Lee Kamentsky]

All very interesting, Gabriel, thank you.

Do you think the specification should include a cursor that iterates 
along the pixels on the perimeter? That would save everyone the trouble 
of reimplementing Chang. It's somewhat trivial for me to include the 
start coordinates in the interface (I'm guessing something like "int [] 
Start(T label);" or "int Start(T label, int dim) instead of or in 
addition to "int Start[X,Y,Z](T label)" to align to the imglib spirit) 
if they're of general use.

The outline pixels can be recovered from a list of pixel coordinates for 
each label by sorting the coordinates by x then y and finding the ones 
without y neighbors on either side, and then sorting by y, then x and 
repeating, although labeling using Chang is clearly more efficient if 
you have a binary thresholding of the image.

We do have a use case in which pixels are multiply labeled (current 
research has images of objects that cross and share pixels) which means 
that the most general case can't use reconstruction of a binary image to 
recover the labeling.

--Lee

On 11/17/2010 4:19 AM, Gabriel Landini wrote:
> I said:
>> Sorry I cannot provide it right now (I have it at work, will do tomorrow)
> The paper is by Chang F, and Chen Ch-J: A Component-Labeling Algorithm Using
> Contour Tracing Technique. Proceedings of the Seventh International Conference
> on Document Analysis and Recognition (ICDAR’03)
>
> http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.95.6330&rep=rep1&type=pdf
>
> I hope this is useful.
> Cheers
>
> Gabriel
>
> _______________________________________________
> ImageJ-devel mailing list
> ImageJ-devel at imagejdev.org
> http://imagejdev.org/mailman/listinfo/imagej-devel