[vtkusers] ComputeTransformedOrigin bug in vtkImageReader.cxx

[Josh Snyder]

The code below reproduces the problem, and is similar to the example 
described further down. The transform supplied to the vtkImageReader 
should reflect the first dimension of the data around x = 0 by 
transforming its -1 0 0 origin to 1 0 0. Instead, the new origin is 0 0 
0. Plugging different origin/dimensions/extent values into this code 
reveals the same offset error every time.

Thanks,
Josh


vtkImageData id
id SetDimensions 1 1 1
id AllocateScalars

vtkImageWriter iw
iw SetFileDimensionality 2
iw SetFileName test.dat
iw SetInput id
iw Write

vtkImageReader ir
ir SetFileDimensionality 2
ir SetFileName test.dat
ir SetDataOrigin -1 0 0
ir SetDataSpacing 1 1 1
ir SetDataExtent 0 0 0 0 0 0

vtkMatrix4x4 m
m SetElement 0 0 -1

vtkTransform t
t SetMatrix m

ir SetTransform t

ir Update
[ir GetOutput] GetOrigin


Mathieu Malaterre wrote:

> Josh,
>
>     Could you provide us with a sample code that reproduce the problem ?
>
> Thanks
> Mathieu
>
> Josh Snyder wrote:
>
>> Making this change fixed my problem. Could someone verify this and 
>> check it in?
>>
>> Josh
>>
>> Josh Snyder wrote:
>>
>>> I've run into a problem using vtkImageReader with the SetTransform 
>>> functionality, and I think it's due to the line of code referenced 
>>> in the post below (line 495 in the current CVS version). It seems to 
>>> me that the older version (w/out the +1) was correct. My 
>>> understanding is that if the Transform specifies an axis reversal, 
>>> the order of the voxels along that axis needs to be reversed, and 
>>> the origin coordinate for that axis needs to be shifted so that the 
>>> voxels originally occurring at negative locations become positive 
>>> and vice versa. The former step is handled by the inverse transform 
>>> of the increments during the vtkImageReaderUpdate2 method, and the 
>>> latter by the code mentioned below. I think the +1 addition is a 
>>> bug. For example, consider the case where the extent in the y 
>>> direction consists of a single voxel situated at y = -1, and the 
>>> Transform specifies that this axis be reversed. This line should 
>>> translate the origin to +1, so that the voxel is newly situated at +1.
>>>
>>> In this case, the variables would be set as follows:
>>> original origin[1] = -1
>>> original spacing[1] = 1
>>>
>>> Transform =
>>> 1 0 0 0
>>> 0 -1 0 0
>>> 0 0 1 0
>>> 0 0 0 1
>>>
>>> yields
>>>
>>> transformedOrigin[1] = 1
>>> transformedSpacing[1] = -1
>>>
>>> transformedExtent[3] = 0
>>> transformedExtent[2] = 0
>>> i = 1
>>>
>>> The current code:
>>> origin[i] = transformedOrigin[i] + 
>>> transformedSpacing[i]*(transformedExtent[i*2+1] -  
>>> transformedExtent[i*2] + 1)
>>> = 1 + (-1 * (0 - 0 + 1)) = 1 - 1 = 0
>>>
>>> The old (correct?) code:
>>> origin[i] = transformedOrigin[i] + 
>>> transformedSpacing[i]*(transformedExtent[i*2+1] -  
>>> transformedExtent[i*2])
>>> = 1 + (-1 * (0 - 0)) = 1 + 0 = 1
>>>
>>> Is my understanding of this correct, or is there something I'm missing?
>>>
>>> Josh
>>>
>>>> Simon,
>>>>
>>>> I checked in your corrections this morning. I also fixed 
>>>> graphics/vtkVolume16Reader. Its logic was
>>>> also flawed.
>>>>
>>>> Thanks,
>>>>
>>>> Bill
>>>>
>>>> -----Original Message-----
>>>> From: Simon Warfield [mailto:warfield at bwh.harvard.edu 
>>>> <mailto:warfield%20at%20bwh.harvard.edu>]
>>>> Sent: Sunday, July 08, 2001 7:09 PM
>>>> To: vtkusers at public.kitware.com 
>>>> <mailto:vtkusers%20at%20public.kitware.com>
>>>> Subject: [vtkusers] bug in vtkImageReader.cxx ?
>>>>
>>>>
>>>>
>>>> I notice that if I read some image data using vtkImageReader and 
>>>> transform
>>>> it to correct for patient orientation using the 
>>>> vtkImagerReader::SetTransform
>>>> function, then the new origin is at an unexpected location.
>>>>
>>>> I get data looking like:
>>>> # vtk DataFile Version 3.0
>>>> vtk output
>>>> BINARY
>>>> DATASET STRUCTURED_POINTS
>>>> DIMENSIONS 256 256 90
>>>> SPACING 0.703125 0.703125 1.5
>>>> ORIGIN -90 -89.2969 -67.5
>>>> CELL_DATA 5787225
>>>> POINT_DATA 5898240
>>>>
>>>> When I expect to get data looking like:
>>>> # vtk DataFile Version 3.0
>>>> vtk output
>>>> BINARY
>>>> DATASET STRUCTURED_POINTS
>>>> DIMENSIONS 256 256 90
>>>> SPACING 0.703125 0.703125 1.5
>>>> ORIGIN -90 -90 -67.5
>>>> CELL_DATA 5787225
>>>> POINT_DATA 5898240
>>>>
>>>>  I think the problem is that line 936 of imaging/vtkImageReader.cxx 
>>>> should not
>>>> read
>>>>          (transformedExtent[i*2+1] -  transformedExtent[i*2]);
>>>> but instead should be:
>>>>          (transformedExtent[i*2+1] -  transformedExtent[i*2]+1);
>>>>
>>>>  This part of vtkImageReader.cxx in the latest vtk nightly is doing 
>>>> a shift of the origin when an axis is reflected, but seems to be 
>>>> using one less
>>>> than the width of the data to compute the new origin. I think the 
>>>> width
>>>> of the data should be used.
>>>>
>>>>  Could someone check this, and make the change to the CVS 
>>>> repository if it
>>>> should be fixed ?
>>>>
>>>> -- 
>>>> Simon Warfield, Ph.D. warfield at bwh.harvard.edu 
>>>> <mailto:warfield%20at%20bwh.harvard.edu> Phone:617-732-7090
>>>> http://www.spl.harvard.edu/~warfield 
>>>> <http://www.spl.harvard.edu/%7Ewarfield>           FAX:  617-582-6033
>>>> Thorn 329, Brigham and Women's Hospital, Harvard Medical School
>>>> Department of Radiology, 75 Francis St, Boston, MA, 02115
>>>>
>>>
>>> _______________________________________________
>>> This is the private VTK discussion list. Please keep messages 
>>> on-topic. Check the FAQ at: <http://public.kitware.com/cgi-bin/vtkfaq>
>>> Follow this link to subscribe/unsubscribe:
>>> http://www.vtk.org/mailman/listinfo/vtkusers
>>
>>
>>
>> _______________________________________________
>> This is the private VTK discussion list. Please keep messages 
>> on-topic. Check the FAQ at: <http://public.kitware.com/cgi-bin/vtkfaq>
>> Follow this link to subscribe/unsubscribe:
>> http://www.vtk.org/mailman/listinfo/vtkusers
>>
>
>
>
> _______________________________________________
> This is the private VTK discussion list. Please keep messages 
> on-topic. Check the FAQ at: <http://public.kitware.com/cgi-bin/vtkfaq>
> Follow this link to subscribe/unsubscribe:
> http://www.vtk.org/mailman/listinfo/vtkusers