detectorcal.fit#

detectorcal.fit.fit_pixel(volume, smoothed, cutoff, i)#

Least squares regression for a single pixel.

Parameters

volume: np.ndarray: The detector-response volume.
smoothed: np.ndarray: The detector-response volume once Gaussian smoothed across xy.
cutoff: scalar: Minimum value in smoothed image at which to include the value in the regression. This is chosen to elminate values that fall outside of the range of linear response.
i: int: index of the pixel in flattened corrdinates (i.e., raveled along x-y plane).

Returns

pixel_fit: scalar: Linear coefficient for the pixel.
i: int: index of the pixel in flattened corrdinates (i.e., raveled along x-y plane).

detectorcal.fit.fit_response(volume, dark=None, save_path=None, sigma=50, mode='nearest', cval=0.0, truncate=4.0, cutoff=300, gpu=False, verbose=False)#

For each pixel in the detector, fit a linear model predicting the true artefact-free detector response to an X-ray beam. For this, a stack of images representing the detector’s response across beam intensities is required. Additionally, a dark current image can be used to correct the dark current for this volume. A Gaussian filter is used to smooth the volume, yielding estimate the true intensity profile of the X-ray beam. The standard deviation of the Gaussian kernel should be chosen to eliminate pixel-wise variations. For each pixel, least-squares regression is then used to determine a coefficient that can be used to map the

Parameters

volume: np.ndarray: Array containing the stack of detector responses.
dark: None or np.ndarray: Default None. Array containing the dark current image or None.
save_path: None or str: Default None. Path to which the fit coefficients should be saved. If None, the coefficients will not be saved to disk. Please save as a tiff, hdf5, or zarr.
sigma: scalar: Default 50. Standard deviation of the Gaussian kernel to be used for smoothing
mode: str: Default ‘nearest’. Determines how the input array is extended at the boarder.
cval: scalar: Default 0.0. Value with which to pad edges when mode ‘constant’ is used.
truncate: float: Default 4.0. truncate filter at this many standard deviations.
gpu: bool: Default determined by whether GPU version is installed (with pip install detectorcal[gpu]). Will gpu acceleration be required (or possible)?
verbose: bool: Default False. Should messages be printed to the console? Will print a link to dask dashboard - this allows you to watch the computation across the workers.
cutoff: scalar: Default 300. Minimum value in smoothed image at which to include the value in the regression. This is chosen to elminate values that fall outside of the range of linear response. Value choice depends on data type and range (e.g., 300 for 12 bit image).

Returns

fit: np.ndarray: Coefficients for each pixel in the detector

References

Croton, L.C., Ruben, G., Morgan, K.S., Paganin, D.M. and Kitchen, M.J.,: 2019. Ring artifact suppression in X-ray computed tomography using a simple, pixel-wise response correction. Optics express, 27(10), pp.14231-14245.

detectorcal.fit.gaussian_smooth(volume: dask.array.core.Array, sigma=50, mode='nearest', cval=0.0, truncate=4.0, gpu=False, verbose=True)#

Apply a 2D Gaussian filter to slices of a 3D volume using dask. Facilitates dask-mediated parallelisation and has optional GPU acceleration.

Parameters

volume: np.ndarray: Array containing a series of x-y planes that require smoothing stacked along z.
sigma: scalar: Standard deviation of the Gaussian kernel to be used for smoothing
mode: str: Determines how the input array is extended at the boarder. For options see: https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.gaussian_filter.html
cval: scalar: Value with which to pad edges when mode ‘constant’ is used.
truncate: float: truncate filter at this many standard deviations.
gpu: bool: Will gpu acceleration be required (or possible)?
verbose: bool: Should messages be printed to the console?

Returns

smoothed: np.ndarray: Smoothed version of the input volume.

detectorcal.fit.parallel_fit(fit, volume, smoothed, cutoff)#: Apply linear regression to each pixel. This is done using a numba accelerated nested for loop.

detectorcal.fit.sequential_fit(volume, smoothed, cutoff)#: Apply linear regression to pixels sequentially.

detectorcal.fit.sequential_gauss(volume: numpy.ndarray, sigma=50, mode='nearest', cval=0.0, truncate=4.0)#: Apply Gaussian smoothing to image planes in sequence.