Add descriptors/features framework about imagej-ops

[ctrueden]

See also ImageJ Trac ticket #1978 for an earlier incarnation of this sort of effort.

[IdealOutage]

Here a more detailed list on what features is already implemented or needs to be implemented. I'm updating this list as we progress. The work is done on the feature-service branch.

Features

1. First Order Statistics

Responsible: @danielseebacher

• Max
• Mean
• Median
• Min
• Quantile
• StdDeviation
• Sum
• Variance
• GeometricMean
• HarmonicMean
• Kurtosis
• Moment1AboutMean
• Moment2AboutMean
• Moment3AboutMean
• Moment4AboutMean
• Percentile
• Skewness
• SumOfInverses
• SumOfLogs
• SumOfSquares

3. Geometric Features

Responsible: @danielseebacher

• Area
• CenterofGravity
• Perimeter
• Eccentricity
• Elongation
• Feret Diameter
• Feret Angle
• Solidity
• Convexity
• Circularity
• Rectangularity
• Roundness
• Aspect Ratio

4. Image Moments 2D

Responsible: @angrauma

• Image Moments
• Central Moments
• Normalized Central Moments
• Hu Moments

5. Tamura Texture Features

• Tamura 2D (@angrauma)
• Tamura 3D

6. Zernike Features

• Zernike 2D (@angrauma)
• Zernike 3D

7. Shape Features

• 2D Fourier Shape Descriptors (@seebacherd)
• 3D Fourier Shape Descriptors

8. Haralick Features

• Haralick 2D (@danielseebacher)
• Haralick 3D (@angrauma)

9. Local Binary Patterns

• Local Binary Patterns 2D (@angraumar)
• Local Binary Patterns 3D

3D: http://lmb.informatik.uni-freiburg.de/papers/download/fe_bu_icpr08.pdf

10. 3D Geometric Features

Responsible: @tibuch

• 3D Geometrical measurements (volume, surface, …) for each labelled object.
• 3D Intensity measurements (mean, integrated density, …)
• 3D Shape measurements (compactness, elongation, …)
• 3D Mesh Measurements after triangulation
• 3D fitting by an ellipsoid and main direction computation
• 3D convex hull

11. 3D Shape Features

Responsible: @tibuch

• http://web.ist.utl.pt/alfredo.ferreira/publications/DecorAR-Surveyon3DShapedescriptors.pdf

12. Gabor Texture Features

Responsible:

13. Image Moments 3D

Responsible:

[dietzc]

we may not forget to offer ops which respect the calibration of the image. In 2D these are especially the geometric features (area for example). In 3D, these are geometric and also (potentially) intensity based features. The calibration of an image can be passed as double[] array. Basically there are two ways to integrate respect calibration in a feature:

• we always provide a calibration as an optional input with default values 1.0.
• we offer two different implementations of the feature, one with calibration one without.

[dietzc]

TODO after reviewing with @ctrueden

• Renaming Def to Default
• get rid of > and RealType<?> completely. Introduce generic types if needed.
• Feature?
  ** Can have some helper methods
• DefaultFeatureService make inner classes private!
• Do I really have to restrict myself on features or can have a "OpCompileService"?!
• OpInfo should be in net.imagej.ops (issue #57). OpArgs ().

[dietzc]

see https://github.com/imagej/imagej-ops/tree/outputop-service

[ctrueden]

@angrauma A little bird told me that Zernike 3D is also working! Anything else that can be checked off the list above?

[dietzc]

@ctrueden haralick 3D, zernike 3D not yet :-) Anyway, @DanielSeebacher, @angrauma: could you add some a marker to the features you are currently working on?

[ctrueden]

@dietzc Oh yeah, right! And it is checked, so I am happy.

[LeeKamentsky]

3D zernike? How are you planning to generalize to multiple dimensions? I found this paper here:
http://cg.cs.uni-bonn.de/aigaion2root/attachments/novotni-2003-3d.pdf
which creates a 3D Zernike-like basis set on top of spherical harmonics (think of electron orbitals if you want to get an idea).
So inquiring minds want to know exactly what you guys might have planned.

[angrauma]

@LeeKamentsky Yes, this is exactly the paper I would like to be oriented towards when implementing 3D Zernike moments. I also want to take a closer look at this paper: http://www.emis.de/journals/HOA/MPE/Volume2012/353406.pdf.
But it seems the idea is very close to the other paper. So it is the plan to realize the 3D Zernike descriptor as an extension of spherical harmonics based descriptors, as it is explained in the paper of Novotni and Klein, but I have not started yet implementing it, so I can not say how I will do it exactly, at the moment.

[dietzc]

another thing we might to work on in the future are quality metrics for images:

e.g. see: http://rsb.info.nih.gov/ij/plugins/mssim-index.html

[maierl]

I would love to see the 3D features! all my data will be in 3D, but my analysis is mostly yet limited to 2D.

[dietzc]

@angrauma (local binary patterns, tamura, 2D Shape etc) / @tibuch (3D geometry, shape etc.) please add the remaining / missing features until September.

[ctrueden]

Status update on this issue?

[dietzc]

I'm implementing the CachedOpEnviroment and FeatureSet these days. The existing implementations of features can then be used. I will clean-up all the branches and open PRs over the next 24h. I will update the issue then.

[ctrueden]

@dietzc Great, thanks!

[dietzc]

... approaching ;-) I will close this issue in favour of issues for the missing feature implementations soon.

[dietzc]

closed in favour of several other issues I just created.