Part 2: finding the center of rotation¶

Finding the CoR is a crucial step in reconstruction. This parameter is probably the most important - without a correct center of rotation, no reconstruction can be done.

Estimating the CoR can be done manually or (almost) automatically.

In this part, we see how to

  • Review different CoR estimation methods available with Nabu
  • Use nabu (command line) to find the CoR
  • Use tomwer (graphical interface) to find the CoR

1 - Center of rotation estimation with Nabu¶

Nabu provides several Center of Rotation (CoR) estimation methods.

See the documentation page for CoR.

For this tutorial, we still use the half-acquisition "bamboo" dataset.

You can generate a configuration file with nabu-config --dataset /where/you/generated/the/bambou_hercules.nx.

1.1 - Where to provide center of rotation (estimation) in configuration file¶

The relevant section in configuration file is [reconstruction].

[reconstruction]
rotation_axis_position = <cor_or_method>

If a scalar value is provided (e.g. 2069.5), then this value will be used for reconstruction.

Otherwise, the parameter is the name of an estimation method (eg. rotation_axis_position = sliding-window).

Note an empty value means the center is half the detector width.

1.2 List of estimation methods¶

Nabu comes with a collection of CoR estimation methods. They work differently (on pairs of radios, sinograms, with correlation or subtraction, etc).

Method name Notes
sliding-window Fast, works well for micro-CT
growing-window Slow, global search
composite-coarse-to-fine Slow, uses an initial estimate. Equivalent to "near" in octave-fasttomo
fourier-angles Fast, needs initial estimate
octave-accurate Fast, only works for centered tomography
sino-sliding-window Same principle as sliding-window, applied on sinogram
sino-growing-window Same principle as growing-window, applied on sinogram

Older, less reliable methods¶

Method name Notes
global Very slow, global search. Works well in nano-CT.
centered Fast, works for centered tomo.
sino-coarse-to-fine Fast, sinogram-based. Needs a 360 degrees scans.

1.3 - Estimators relying on an initial estimate¶

ESRF Bliss acquisition files come with an initial estimate of the center of rotation. This CoR estimate is done when initially aligning the beamline setup.

This value can be obtained by dividing the "yrot" motor position (usually instrument/positioners/yrot in Bliss files) with the pixel size.

In the NX file, this information can be found in instrument/detector/estimated_cor_from_motor, in pixel units, offset from the middle of the detector.

By default, nabu will automatically look in the NX file for the estimated_cor_from_motor field.

Caution

  • In the NX file, estimated_cor_from_motor is a position (in pixel units) from middle of the detector
  • In nabu configuration rotation_axis_position is a position (in pixel units) from the left-most pixel of the detector

The following estimators should use an initial estimate - they work well, but will have poor performance without a good-enough initial estimate:

  • composite-coarse-to-fine (aka "near")
  • fourier-angles

The following estimators can use an initial estimate - they still have decent performance if not:

  • sliding-window (and sino-sliding-window)

All the other estimators do not use an initial estimate.

Exercise¶

  • Use sliding-window on the "bamboo" dataset - it should correctly find the CoR.
  • Set rotation_axis_position=2069.5 (this will force nabu to use this value), run the reconstruction of your favourite slice and compare with previously.

1.4 - Advanced options¶

Most estimators can be tuned with advanced options. In nabu, the default options try to be good for most cases.

In the configuration file, these options are provided in the field cor_options.

For example, to use sliding-window and look to the right side of the detector:

[reconstruction]
rotation_axis_position = `sliding-window`
cor_options = side="right"

The syntax of cor_options is admittedly cumbersome:

  • Don't forget quotes for side="right"
  • Parameters are separated by semicolon: side="right"; low_pass=1; high_pass=20

Exercise¶

  • Set the config file as below and run the reconstruction.
[reconstruction]
rotation_axis_position = `sliding-window`
cor_options = side=2069.5
  • Compare the obtained COR value obtained. In this case, nabu passes that value to the sliding-window algo as a first guess.

1.5 - Reconstructing one slice with several COR values at once¶

One can also reconstruct one slice at several COR positions at once with

nabu-multicor nabu.conf middle 2049:2069:1

This will reconstruct the middle slice at all COR values from 2049 to 2069, with a step of 1 pixel. All the other options are taken from the nabu.conf file.

2 - Using tomwer "semi-automatic" rotation axis finder¶

See the dedicated notebook