[Insight-users] Problems about ImageBoundaryFacesCalculator

Miller, James V (Research) millerjv at crd.ge.com
Fri, 16 Apr 2004 09:35:06 -0400


This message is in MIME format. Since your mail reader does not understand
this format, some or all of this message may not be legible.

------_=_NextPart_001_01C423B7.A139E624
Content-Type: text/plain

The boundary faces calculator attempts to decompose an image to
NON-overlapping regions.
 
However, there is currently a bug in the implementation where under certain
radii and image size combinations, a few pixels can appear in multiple
regions. (In extreme cases where one image dimension is less than the radii,
a large percentage of the pixels can appear in multiple images).  This is
not only a problem for performance reasons, it also creates the potential
for errors the a filter is run on multiple processors (i.e. several threads
could attempt to set the same output pixel at the same time).
 
In most circumstances, however, the overlap is zero or simply a few pixels.
 
I attempted to fix this prior to the 1.4 release but had a problem with my
implementation on linux.  I will re-visit this issue.  The behavior should
be that each pixel should only appear in one region.  The set of regions
outputted with always be ordered as the first region wil lhave no boundary
conditions (but the region may be empty) and the remainder of the regions
will all have boundary conditions.
 
Jim

-----Original Message-----
From: Yan Shang [mailto:yan.shang at ibt.uni-karlsruhe.de]
Sent: Friday, April 16, 2004 6:52 AM
To: insight-users at itk.org
Subject: [Insight-users] Problems about ImageBoundaryFacesCalculator



dear Luis, 


I have the following question: 


Using NeighborhoodAlgorithm::ImageBoundaryFacesCalculator will decomposed
the input image into overlapped regions, that means some pixels can belong
to several boundary regions ( i.e., for 3D, radius=3, the pixels ( in the
upper left 3*3*3 region) will belong at the same time to the boundary
regions in x, y, z direction.) 


If we use the above mentioned decomposed regions of the input image in a
filter, the pixels belonging to several regions will be calculated
repeatedly, this will cause a great computational burden if the image and
radius are both large. 


Am I right in this point? 


yan 


------_=_NextPart_001_01C423B7.A139E624
Content-Type: text/html

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=us-ascii">
<TITLE></TITLE>

<META content="MSHTML 6.00.2800.1276" name=GENERATOR></HEAD>
<BODY>
<DIV><SPAN class=053012813-16042004><FONT color=#0000ff size=2>The boundary 
faces calculator attempts to decompose an image to NON-overlapping 
regions.</FONT></SPAN></DIV>
<DIV><SPAN class=053012813-16042004><FONT color=#0000ff 
size=2></FONT></SPAN>&nbsp;</DIV>
<DIV><SPAN class=053012813-16042004><FONT color=#0000ff size=2>However, there is 
currently a bug in the implementation where under certain radii and image size 
combinations, a few pixels can appear in multiple regions. (In extreme cases 
where one image dimension is less than the radii, a large percentage of the 
pixels can appear in multiple images).&nbsp; This is not only a problem for 
performance reasons, it also creates the potential for errors the a filter is 
run on multiple processors (i.e. several threads could attempt to set the same 
output pixel at the same time).</FONT></SPAN></DIV>
<DIV><SPAN class=053012813-16042004><FONT color=#0000ff 
size=2></FONT></SPAN>&nbsp;</DIV>
<DIV><SPAN class=053012813-16042004><FONT color=#0000ff size=2>In most 
circumstances, however, the overlap is zero or simply a few 
pixels.</FONT></SPAN></DIV>
<DIV><SPAN class=053012813-16042004><FONT color=#0000ff 
size=2></FONT></SPAN>&nbsp;</DIV>
<DIV><SPAN class=053012813-16042004><FONT color=#0000ff size=2>I attempted to 
fix this prior to the 1.4 release but had a problem with my implementation on 
linux.&nbsp; I will re-visit this issue.&nbsp; The behavior should be that each 
pixel should only appear in one region.&nbsp; The set of regions outputted with 
always be ordered as the first region wil lhave no boundary conditions (but the 
region may be empty) and the remainder of the regions will all have boundary 
conditions.</FONT></SPAN></DIV>
<DIV><SPAN class=053012813-16042004><FONT color=#0000ff 
size=2></FONT></SPAN>&nbsp;</DIV>
<DIV><SPAN class=053012813-16042004><FONT color=#0000ff 
size=2>Jim</FONT></SPAN></DIV>
<BLOCKQUOTE dir=ltr style="MARGIN-RIGHT: 0px">
  <DIV class=OutlookMessageHeader dir=ltr align=left><FONT face=Tahoma 
  size=2>-----Original Message-----<BR><B>From:</B> Yan Shang 
  [mailto:yan.shang at ibt.uni-karlsruhe.de]<BR><B>Sent:</B> Friday, April 16, 2004 
  6:52 AM<BR><B>To:</B> insight-users at itk.org<BR><B>Subject:</B> [Insight-users] 
  Problems about ImageBoundaryFacesCalculator<BR><BR></FONT></DIV>
  <P>dear Luis, </P><BR>
  <P>I have the following question: </P><BR>
  <P>Using <FONT 
  face="Courier New">NeighborhoodAlgorithm::</FONT>ImageBoundaryFacesCalculator 
  will decomposed the input image into overlapped regions, that means some 
  pixels can belong to several boundary regions ( i.e., for 3D, radius=3, the 
  pixels ( in the upper left 3*3*3 region) will belong at the same time to the 
  boundary regions in x, y, z direction.) </P><BR>
  <P>If we use the above mentioned decomposed regions of the input image in a 
  filter, the pixels belonging to several regions will be calculated repeatedly, 
  this will cause a great computational burden if the image and radius are both 
  large. </P><BR>
  <P>Am I right in this point? </P><BR>
  <P>yan </P></BLOCKQUOTE></BODY></HTML>

------_=_NextPart_001_01C423B7.A139E624--