<html>
<body>
Hi all,<br><br>
Three weeks ago i posted the following mail (see also below this
email):<br><br>
<a href="http://public.kitware.com/pipermail/insight-users/2006-October/019744.html" eudora="autourl">http</a><a href="http://public.kitware.com/pipermail/insight-users/2006-October/019744.html" eudora="autourl">://public.kitware.com/pipermail/insight-users/2006-October/019744.html<br><br>
</a>concerning a possible bug in the
itkBSplineInterpolationWeightFunction. Has anybody looked at this maybe?
Should I add it as a bug entry in the bug tracker, or is it not a bug
maybe?<br><br>
See also a followup mail that i sent: <br>
<a href="http://public.kitware.com/pipermail/insight-users/2006-October/019747.html" eudora="autourl">http://public.kitware.com/pipermail/insight-users/2006-October/019747.html</a><br><br>
Kind regards!<br>
Stefan.<br><br>
<br><br>
<br>
<blockquote type=cite class=cite cite><font size=3>Hi,<br><br>
I think i found a bug in the implementation of the
BSplineInterpolationWeightFunction.<br><br>
In the Evaluate function, at lines 146-151, the startIndex of the support
region is computed, given a continuous input index (denoted by
'index'):<br><br>
// Find the starting index of the support region<br>
for ( j = 0; j < SpaceDimension; j++ )<br>
{<br>
startIndex[j] = static_cast<typename
IndexType::IndexValueType>(<br>
BSplineFloor( index[j] -
static_cast<double>( SplineOrder / 2 ) ) );<br>
}<br><br>
<br>
If I understand well, for a zero order bspline, this formula should
result in a 'round' operation. This is not the case with the formula
above:<br><br>
suppose the continuous input index = 0.9 and the SplineOrder = 0.<br>
then:<br><br>
startindex = floor( 0.9 - 0/2 ) = floor (0.9) = 0.<br><br>
Consequently, in line 161, where the bsplinekernel weight is evaluated,
the result is zero.<br>
BsplineZeroOrder( inputindex - startindex ) = BSplineZeroOrder( 0.9 ) =
0.0.<br>
Which must be wrong, since for a zeroth order bspline the weight value
should be nonzero everywhere in its support region.<br><br>
For splineorder=1 the formula works. For splineorder=2 it goes wrong
again and for splineorder=3 it works. So, it's wrong for even spline
orders.<br><br>
I suggest changing it to:<br><br>
startIndex = floor( inputIndex + bsplineoffset
)<br><br>
where<br><br>
bsplineOffset = 0.5 -
static_cast<double>(SplineOrder) / 2.0 ;<br><br>
which results in bsplineOffset = 1/2, 0, -1/2, or -1 respectively for
order 0, 1, 2, and 3.<br><br>
In the case of zero order bspline this obviously gives the desired round
operation (floor(input+0.5)).<br>
In general: for odd spline orders it works the same as the current
formula, but for even orders the result is different.<br><br>
Please let me know if I made a mistake, or if it's really a
bug.<br><br>
Kind regards,<br>
Stefan.<br>
</font></blockquote></body>
<br>
</html>