Page MenuHomePhabricator
Create Task
Maniphest T8078

Offset-Bug in ImageStatisticsCalculator
Closed, ResolvedPublic
Actions

Description

When using mitkImageStatisticsCalculator with a PlanarFigure to define a ROI, there is a severe offset bug.
If the right border of a pixel is included in the ROI, it is completely used for masking.

That means if I have a ROI that encircles one pixel but does not cross the right border to the next pixel I have no statistics.

If this ROI contains 99% of the pixel on the right of the first pixel, this second pixel will NOT be considered in the statistics.

If this ROI contains 1% of the pixel on the left side (only the righternmost 1%), this pixel is completely considered by the statistics.

The most accurate way to fix this would be to make the statistics calculation work correctly on sub-pixel level.
However, this would mean immense workload and will not be done in the near future.

The quick-fix / improvement would be to use a pixel for statistics calculation as soon as the planarFigure crosses the center of the pixel. This would fix the false behaviour mentioned above.

Related Objects
Search...

Event Timeline

engelm added projects: MITK, Bugzilla.May 10 2011, 4:19 PM
engelm added a subscriber: engelm.

Screenshot showing the false behaviour

screenshot.png (1×1 px, 162 KB)

engelm added a comment.May 10 2011, 4:27 PM

[d57e8f]: Merge branch 'bug-8078-imagestatistics-planarfigure-offset'

  • bug-807

Merged commits:

2011-05-10 16:22:43 Markus Engel [15f52e]
removing coordinate-correction to fix offset bug for planarfigure masking

engelm added a comment.May 10 2011, 4:49 PM

Removed two lines of code in mitkStatisticsCalculator that were responsible for the 0.5 pixel offset in the creation of a masking image using a PlanarFigure.

Statistics calculation works as expected now...

engelm added a comment.May 12 2011, 3:20 PM

Needs to be reopened for real fix of problem

engelm added a comment.May 12 2011, 3:42 PM

[4196b3]: Merge branch 'bug-8078-imagestatistics-planarfigure-offset'

  • bug-807

Merged commits:

2011-05-12 15:39:36 Markus Engel [1cede8]
introduced new pipeline to create imagemask from PlanarFigure;

The old pipeline has been replaced by a simple filter vtkLassoStencilSource that
creates a vtkImageStencil from a polygon. This new approach works correctly
for all kinds of intersections of pixels.
The two new classes will be released in VTK 5.8, that why we have them here
temporarily.

2011-05-12 15:28:24 Daniel Maleike [fa3a1d]
Build a switch to use two VTK 5.8 classes even when using VTK 5.6 (with VTK 5.8, use the real VTK classes)

2011-05-12 15:17:31 Markus Engel [92bad7]
adding classes from VKT 5.8 needed for statistics calculation

engelm added a comment.May 12 2011, 4:49 PM

When intersecting pixels at an off angle the behaviour of the existing filter-pipeline was rather unpredictable. Some pixels were considered while other pixels that were intersected in the same way were not.

VTK 5.8 will introduce a new class vtkLassoStencilSource that creates a vtkImageStencil from the points of a polygon (exactly what we want).

I copied the two classes that were needed ( vtkLassoStencilSource & vtkImageStencilRaster ) to mitkExt and introduced '#ifdef's to make sure the real vtk classes are used as soon as vtk 5.8 is used.

The resulting pipeline is shorter, the readability is improved and the results are a lot more accurate (has been verified by a Unit-test that is developed under T8077)

engelm added a comment.May 13 2011, 2:00 PM

reopen for review-stuff

engelm added a comment.May 13 2011, 2:01 PM

[b9c40c]: Merge branch 'bug-8078-imagestatistics-planarfigure-offset'

  • bug-807

Merged commits:

2011-05-13 13:55:53 Markus Engel [7aef88]
removed commented code

engelm added a comment.May 16 2011, 8:53 AM

reopen for rebase of branch

engelm added a comment.May 18 2011, 9:48 AM

Branch is everywhere we need it, can be closed...

MITK · MIC · IMSY · DKFZ