A few weeks ago in a discussion of calculating lengths of long narrow
objects I suggested skeletonizing and the counting the pixels. This works
fine for straight objects. However, as the objects curve more and more,
this becomes inaccurate. We ran a simulation with a random function of 1
+ a random function of zero to 0.414 multiplied by the pixel counts
compared to measurements of objects by the perimeter method and found that
the perimeter measurement is far more accurate. For the objects we're
measuring, the difference between perimeter and random function multiplied
by pixel counts for the skeletonized objects never exceeds 5% error with
hand traced measurements falling in between.
_________________________________________
Michael Cammer
http://www.aecom.yu.edu/aif/