Skip to content

nabu.thirdparty.tomocupy_remove_stripe

source module nabu.thirdparty.tomocupy_remove_stripe

This file is a "GPU" (through cupy) implementation of "remove_all_stripe". The original method is implemented by Nghia Vo in the algotom project: https://github.com/algotom/algotom/blob/master/algotom/prep/removal.py The implementation using cupy is done by Viktor Nikitin in the tomocupy project: https://github.com/tomography/tomocupy/blame/main/src/tomocupy/remove_stripe.py then moved to https://github.com/tomography/tomocupy/blob/main/src/tomocupy/processing/remove_stripe.py

For now we can't rely on off-the-shelf tomocupy as it's not packaged in pypi, and compilation is quite tedious.

License follows.

Classes

  • DWTForward — Performs a 2d DWT Forward decomposition of an image

  • DWTInverse — Performs a 2d DWT Inverse reconstruction of an image

Functions

  • afb1d — 1D analysis filter bank (along one dimension only) of an image

  • sfb1d — 1D synthesis filter bank of an image Array

  • remove_stripe_fw — Remove stripes with wavelet filtering

  • remove_stripe_ti — Remove stripes with a new method by V. Titareno

  • remove_all_stripe — Remove all types of stripe artifacts from sinogram using Nghia Vo's approach :cite:Vo:18 (combination of algorithm 3,4,5, and 6).

  • remove_all_stripe_sinos — Same as remove_all_stripe(), but acting on sinograms

  • remove_all_stripe_pycuda — Nabu interface to tomocupy "remove_all_stripe". Processing is done in-place to save memory, meaning that the content of "array" will be overwritten.

source afb1d(x, h0, h1='zero', dim=-1)

1D analysis filter bank (along one dimension only) of an image

Parameters

  • x (array) : 4D input with the last two dimensions the spatial input

  • h0 (array) : 4D input for the lowpass filter. Should have shape (1, 1, — h, 1) or (1, 1, 1, w)

  • h1 (array) : 4D input for the highpass filter. Should have shape (1, 1, — h, 1) or (1, 1, 1, w)

  • dim (int) - dimension of filtering. d=2 is for a vertical filter (called — column filtering but filters across the rows). d=3 is for a horizontal filter, (called row filtering but filters across the columns).

Returns

  • lohi : lowpass and highpass subbands concatenated along the channel — dimension

source sfb1d(lo, hi, g0, g1='zero', dim=-1)

1D synthesis filter bank of an image Array

source class DWTForward(wave='db1')

Performs a 2d DWT Forward decomposition of an image

Parameters

  • wave : str — Which wavelet to use.

Methods

  • apply — Forward pass of the DWT.

source method DWTForward.apply(x)

Forward pass of the DWT.

Parameters

  • x : array — Input of shape :math:(N, C_{in}, H_{in}, W_{in})

Returns

  • (yl, yh) — tuple of lowpass (yl) and bandpass (yh) coefficients. yh is a list of scale coefficients. yl has shape :math:(N, C_{in}, H_{in}', W_{in}') and yh has shape :math:list(N, C_{in}, 3, H_{in}'', W_{in}''). The new dimension in yh iterates over the LH, HL and HH coefficients.

Note

:math:H_{in}', W_{in}', H_{in}'', W_{in}'' denote the correctly downsampled shapes of the DWT pyramid.

source class DWTInverse(wave='db1')

Performs a 2d DWT Inverse reconstruction of an image

Parameters

  • wave : str — Which wavelet to use.

Methods

source method DWTInverse.apply(coeffs)

Parameters

  • coeffs : yl, yh — tuple of lowpass and bandpass coefficients, where: yl is a lowpass array of shape :math:(N, C_{in}, H_{in}', W_{in}') and yh is a list of bandpass arrays of shape :math:list(N, C_{in}, 3, H_{in}'', W_{in}''). I.e. should match the format returned by DWTForward

Returns

  • Reconstructed input of shape — math:(N, C_{in}, H_{in}, W_{in})

Note

:math:H_{in}', W_{in}', H_{in}'', W_{in}'' denote the correctly downsampled shapes of the DWT pyramid.

source remove_stripe_fw(data, sigma, wname, level)

Remove stripes with wavelet filtering

source remove_stripe_ti(data, beta, mask_size)

Remove stripes with a new method by V. Titareno

source remove_all_stripe(tomo, snr=3, la_size=61, sm_size=21, dim=1)

Remove all types of stripe artifacts from sinogram using Nghia Vo's approach :cite:Vo:18 (combination of algorithm 3,4,5, and 6).

Parameters

  • tomo : ndarray — 3D tomographic data.

  • snr : float — Ratio used to locate large stripes. Greater is less sensitive.

  • la_size : int — Window size of the median filter to remove large stripes.

  • sm_size : int — Window size of the median filter to remove small-to-medium stripes.

  • dim : {1, 2}, optional — Dimension of the window.

Returns

  • ndarray — Corrected 3D tomographic data.

source remove_all_stripe_sinos(sinos, snr=3, la_size=61, sm_size=21, dim=1)

Same as remove_all_stripe(), but acting on sinograms

source remove_all_stripe_pycuda(array, layout='radios', device_id=0, **kwargs)

Nabu interface to tomocupy "remove_all_stripe". Processing is done in-place to save memory, meaning that the content of "array" will be overwritten.

Parameters

  • array : pycuda.GPUArray — Stack of radios in the shape (n_angles, n_y, n_x), if layout == "radios" Stack of sinos in the shape (n_y, n_angles, n_x), if layout == "sinos".

Other Parameters

See parameters of 'remove_all_stripe