ri-imaging/radontea

python library for tomographic image reconstruction

radontea's Issues

new image

Hello,
How can I make radon transfrom to custom image with radontea?

Refactoring

• remove leading "_" in submodule names

Bug in `utils.compute_angle_weights_1d`

I believe this function should return a flat line in case all the angles in theta are equidistant.

This is true in this example:

theta = np.arange(0., 360., 1.)
>> [0.  1.  2.  ...  357.  358.  359.]
weights = utils.compute_angle_weights_1d(np.deg2rad())
>> [1.  1.  1. ... 1.  1.  1.]

But there seems to be a problem in the following cases:

1) Not a full circle

But it is not the case on the edge angles (first and last in the vector) if we do not do a full circle:

theta = np.arange(0., 136., 1.)  
>> [0.  1.  2. ...  133.  134.  135.]
weights = utils.compute_angle_weights_1d(np.deg2rad())
>> [17.37777778  0.75555556  0.75555556  ...  0.75555556  0.75555556 17.37777778]

# theta = np.arange(-90., 46., 1.)  # This theta vector produces the same result

2) Edges lying on the same line

And then there is another problem, if the edge angles lie on the same line through the origin of the unit circle:

theta = np.arange(-90., 91., 1.)
>> [-90.  -89.  -88.  ...  88.  89.  90.]
weights = utils.compute_angle_weights_1d(np.deg2rad())
>> [0.50277778 1.00555556 1.00555556  ...  1.00555556 1.00555556  0.50277778]

# This theta vector produces the same result as above
theta = np.arange(-80., 101., 1.) 

3) Edges are the same angle

Also, when the edge angles are the same, we get a 0 weight on the 90°, which does not make sense IMO:

theta = np.arange(-90., 271., 1.)
>>
weights = utils.compute_angle_weights_1d(np.deg2rad())
>> [-90.  -89.  -88.  ...  89.  90.  91.  ...  268.  269.  270.]
>> [1.00277778  1.00277778  1.00277778  ...  1.00277778  0.  1.00277778  ...  1.00277778  1.00277778  1.00277778]

Summary

Either this is not a correct implementation, or I do not understand the intentions with this function.
When I try to reconstruct a tomogram from the projections recorded at the usual tilt angles (-60°, 60°), the two edge tilts get extremely big weights and the reconstruction looks terrible. Thanks for looking at this.

GitHub Actions: Tests fail inconsistently

Original: 2D Fourier mapping test fails on Windows and macOS
(which might be related - the tests could just be bad)

Add example to documentation

convert the python -m stuff to a separate example and add it to the docs.