[Insight-users] bug in itkImageSpatialObject

[Stefan Klein]

Hi,

I believe I found a bug in the 
ImageSpatialObject::ComputeLocalBoundingbox() method.

The bounding box is wrongly computed in case of rotated images. Attached 
is a test program that produces the bug and tests a fix. A 100x100 image 
is created, with a rotation matrix of 45 degrees (pi/4 rad) as direction 
cosines. Below are the resulting bounding box min/max coordinates:

ITK:
min: [0, 0]
max: [70.7107, 70.7107]
FIX:
min: [-70.7107, 0]
max: [70.7107, 141.421]

ITK = the itk implementation
FIX = my proposed fix (itkImageSpatialObject2).

The bounding box of a rotated square of 100x100 can never become smaller 
than 100x100, so the ITK result must be wrong.

By changing the variable 'factor' in the test program you will see that 
with even values both versions give a correct result, but with odd 
values the ITK version is faulty.

The cause of the failure lies in itkImageSpatialObject.txx lines 
256-260. By setting the minimum and maximum first, these point will 
never be considered to be maximum or minimum, respectively.

Kind regards,
Stefan.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: testspatobbb.zip
Type: application/x-zip-compressed
Size: 6106 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20090828/6cc306dc/attachment.bin>