The author describe some parameters computed on shapes in a label image and, for some of the parameters, on the intensity values inside the image. The mathematics behind the parameters are detailed as well as the way to use the provided code.

The source code is provided as well as the input images.

I've been able to download, build and run the code with the test program. I haven't been able to validate the result produced, as the author doesn't provide the output data (parameter values and rotated images). I would suggest checking the parameter values and the output image in the test.

ITK all the way!

The code is usable without any problem, and very clearly documented both in the article and in the code. Using it is a matter of minutes.

The code is very well commented and thus easy to read. It builds without problem on my mac and linux platforms.

Caracterizing the shape of a collection of objects is a very common task after the segmentation. It can be useful in almost all the domains where a segmentation is performed.
However, the lack on usage of the image spacing may be a problem in some field, like confocal microscopy, where a different spacing is almost always used on the z axis.

This contribution fills an important lack in ITK, and gives strong background on the parameters computed. However, the simple architecture chosen to store the computed values makes it difficult to use them for anything else than making measures. I beleive that this contribution would be a great improvement for another one I made: http://hdl.handle.net/1926/584 , because in that contribution, the parameters are usable for more complex manipulations like opening by attribute, relabeling, etc. I hope we'll be able to work on the integration of those to contributions in a near future!

I appreciate the details explained in the paper about the image moment computation, especially the addition of the number 1/12. Do you think that you could extend the demonstration with the usage of the physical location, and in particular of the spacing?

Finally, you write in the paper that the elongation can't be extended to ND - I think it can, as well as the flatness (which is different of the elongation when D>2). In my contribution, I used the following definitions, which seem to be used for a long time in particule characterization fields:

    elongation = vcl_sqrt(principalMoments[ImageDimension-1] / principalMoments[ImageDimension-2]);
    flatness = vcl_sqrt(principalMoments[1] / principalMoments[0]);

This paper presents a filter for mesuring geometric information of labeled objects.

The authors have provided source code and input images with this paper.

The code compiles fine on MSVC 2008 express and runs well.

This code uses ITK.

This code proposes a contribution to ITK.

The inside clas LabelGeometry should probably be implemented in the .txx class, especially since the implementation is extensive.

Some of the indentations are wrong. Other than that the code is well documentated and fits ITK coding style.

Great paper, very well written and detailed.

Following common image segmentation tasks it is often desireable to compute geometrical features of the objects for image analysis and pattern recognition.  The existing itkLabelStatisticsImageFilter allows for statistical measurements of labeled regions in the intensity image, however, there was a need for geometrical measurements of the label image itself.  The authors created the itkLabelGeometrImageFilter to address this need, and provide a filter that can be easily expanded to calculate new features as the need arrises.

The authors build upon existing ITK filters in their work.

The source code is provided and it follows the itk filter pattern, making it very usable.  I was able to download and integrate the code into my project in a matter of minutes.

The code is easy to follow, it uses the standard ITK coding style, and I have tested it on several platforms using a variety of compilers.

 I had some experience with processing  labeled images using this filter.

 It is a very neat filter that wraps many useful statistics features from labeled regions or objects. It uses hash map to  store and look up  the features  for each labeled region, which makes the filter computationally efficient.

The code itself is very well documented and structured. 

Summary:

The authors described a filter which can be used to determine the properties (volumes, centroids, ...) of regions in an labeled ND image.  It is similar to MATLAB’s "regionprops" command.

Open Science:

The authors provided the source code and a test image.  This was enough to replicate their work.

Reproducibility:

I was able to reproduce the authors’ work.  The test code compiled and ran without any modifications in Visual Studio 2005 on Windows XP.  I integrated the code into a project I am working on.  I used this filter to find the centroids, volumes and eccentricities of regions in a 2D binary image.

Use of Open Source Software:

The authors used ITK, an open source library.

Open Source Contributions:
The authors provided the filter’s source code, test code and a test image.

Code Quality:

The test code (itkLabelGeometryImageFilterTest.cxx) and source was easy to follow and understand.

Additional Comments:
This was very useful in translating MATLAB code which used "regionprops" to C++.