[Insight-users] Neighborhood Connected region growth algorithm description on ITK software guide

[Quan Chen]

I believe that there is something wrong with the explanation of Neighborhood
Connected region growth algorithm (ITK Software Guide, Aug. 21, 2003, page
319).  
 
The explanation says first "The NeighborhoodConnectedImageFilter, on the
other hand, will only accept a pixel if all its neighbors have intensities
that fit in the interval."
 
Then it claim "This filter is more resistant to the presence of noise in the
input image and may not require any initial filtering to smooth the image."
 
This is wrong.  If there are a noise point which the value exceed the
pre-defined intensity interval, the old connected threshold region growth
algorithm only exclude that point from the region, whereas the neighborhood
connected region growth algorithm will exclude all points in the neighbor of
that noise point. Therefore, the noise will have a much bigger impact on the
neighborhood connected region growth algorithm than connected threshold
region growth method.  For the connected threshold region growth method, you
could do a dilation and erosion at last stage to solve the noise problem.
 
Also, the manual said " The reason for considering the neighborhood
intensities instead of only the current pixel intensity is that isolated
pixels are less likely to be accepted in the region. This can be seen as a
preemptive mathematical morphology operation that is similar to using the
ConnectedThresholdImageFilter and then applying a combination of erosion and
dilation with a structuring element of the same radius used for the
neighborhood provided to the NeighborhoodConnected-ImageFilter."
 
This is also wrong, as my last point, the noise in neighborhood connected
region growth method will affect all its neighbors, therefore, this method
will similar to using the ConnectedThresholdImageFilter and then applying
ONLY the erosion operation (and choose only the region connected to the
seed).  This is also evident when you compare Fig. 9.1 to Fig. 9.2.
 
So I believe either the explanation of the algorithm is incorrect, or the
implementation of the algorithm is incorrect.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20050107/2809bd39/attachment-0001.html