[Insight-users] The Hough transform and world/pixel coordinates

[Markus Neuner]

Hi Alberto Gomez,

When you take a look at the source code of
"Insight/Code/BasicFilters/itkHoughTransform2DLinesImageFilter.txx"
starting at line 298 you can see how the points are calculated:

...
        LineType::PointListType list; // insert two points per line
...
        double radius = it_input.GetIndex()[0];
        double teta   =
((it_input.GetIndex()[1])*2*nPI/this->GetAngleResolution())-nPI;
        double Vx = radius*vcl_cos(teta );
        double Vy = radius*vcl_sin(teta );
        double norm = vcl_sqrt(Vx*Vx+Vy*Vy);
        double VxNorm = Vx/norm;
        double VyNorm = Vy/norm;
...
          LinePointType p;
          p.SetPosition(Vx,Vy);
          list.push_back(p);
          p.SetPosition(Vx-VyNorm*5,Vy+VxNorm*5);
          list.push_back(p);
...

Furthermore the points are of type "itk::LineSpatialObjectPoint<
TPointDimension >" whith "typedef Point
<http://www.itk.org/Doxygen/html/classitk_1_1Point.html>< double,
TPointDimension > PointType".
<http://www.itk.org/Doxygen/html/classitk_1_1LineSpatialObjectPoint.html#993f2f73923d9e4fc69f18ca6ac95ea4>

As you can see the coordinates are of type double.

The easiest solution is to construct a
 itk::ContinuousIndex
<http://www.itk.org/Doxygen/html/classitk_1_1ContinuousIndex.html>
(should be defined in the ImageType::ContiunuousIndexType)
and pass it to
TransformContinuousIndexToPhysicalPoint (const ContinuousIndex<
TCoordRep, VImageDimension > &        index, Point< TCoordRep,
VImageDimension > &  point ).

I hope this helps.

Best regards,
Markus



Gomez, Alberto wrote:
> Hi all,
>
> I am using the itk::HoughTransform2DLinesImageFilter  to extract the
> lines from an image for a certain application. The point is that I
> don't want to draw the lines but to get the lines parameters
> (direction and a point ) for further computations. I realised that,
> following the example from the software guide, the vector and the line
> are in pixel coordinates and not in real world coordinates. Can anyone
> confirm this?
>
> So, what I did is to use the TransformIndexToPhysicalPoint() method,
> but it does not work properly (mi lines are slightly deviated). My
> assumption is that the extracted points are in pixel coordinates but
> aren't integers. Because of that, the TransformIndexToPhysicalPoint()
> method rounds them and then computes the physical coordinates, which
> yields in the inaccuracy I realized.
>
> The solution I found was to get two points, in pixel coordinates, very
> far from each other, so that the inaccuracy of the conversion is
> smaller. Nevertheless, this is not clean. Does anyone know a way to
> get the line parameters accurately in physical coordinates? I guess
> that the Hough map (output of the filter) is (r, theta) in real
> physical coordinates, but I don't know how to extract the maximums in
> the same way that getLines() does (using the disc, etc).
>
> Thank you,
> Alberto
> ||
>
> -- 
>
> Alberto Gómez
>
> /Division of Imaging Sciences
> The Rayne Institute
> 4th Floor, Lambeth Wing
> St Thomas' Hospital
> London SE1 7EH /
>
> phone: +44 (0) 20 718 88364
> email: alberto.gomez at kcl.ac.uk <mailto:alberto.gomez at kcl.ac.uk>
>
> ------------------------------------------------------------------------
>
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Kitware offers ITK Training Courses, for more information visit:
> http://www.kitware.com/products/protraining.html
>
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
>