nd.filters package

The main use of image filters is for noise reduction. This module implements several such filters, all of which are designed to work in an arbitrary number of dimensions.

nd.filters._expand_kernel(kernel, kernel_dims, new_dims)[source]

Reshape a kernel spanning some dimensions to cover a superset of dimensions.

Parameters

  • kernel (ndarray) – An n-dimensional kernel.

  • kernel_dims (tuple) – The dimensions corresponding to the kernel axes.

  • new_dims (tuple) – The dimensions of the dataset to which the kernel needs to be applied. Must be a superset of kernel_dims.

Returns

The reshaped kernel.

Return type

ndarray

Raises

  • ValueError – Will raise a ValueError if the dimensions of the kernel don’t match kernel_dims.

  • ValueError – Will raise a ValueError if new_dims is not a superset of kernel_dims.

class nd.filters.BoxcarFilter(dims=('y', 'x'), w=3, **kwargs)[source]

Bases: nd.filters.ConvolutionFilter

A boxcar filter.

Parameters

  • dims (tuple of str, optional) – The dimensions along which to apply the filter (default: (‘y’, ‘x’)).

  • w (int) – The width of the boxcar window. Should be an odd integer in order to ensure symmetry.

  • kwargs (dict, optional) – Extra keyword arguments passed on to scipy.ndimage.filters.convolve.

apply(ds, inplace=False, *, njobs=1)[source]

Apply the filter to the input dataset.

Parameters

  • ds (xarray.Dataset) – The input dataset

  • inplace (bool, optional) – If True, overwrite the input data inplace (default: False).

  • njobs (int, optional) – Number of jobs to run in parallel. Setting njobs to -1 uses the number of available cores. Disable parallelism by setting njobs to 1 (default).

Returns

The filtered dataset

Return type

xarray.Dataset

dims = ()
kwargs = {}
per_variable = True
supports_complex = True
class nd.filters.ConvolutionFilter(dims=('y', 'x'), kernel=None, **kwargs)[source]

Bases: nd.filters.Filter

Kernel-convolution of an xarray.Dataset.

Parameters

  • dims (tuple, optional) – The dataset dimensions corresponding to the kernel axes (default: (‘y’, ‘x’)). The length of the tuple must match the number of dimensions of the kernel.

  • kernel (ndarray) – The convolution kernel.

  • kwargs (dict, optional) – Extra keyword arguments passed on to scipy.ndimage.filters.convolve.

apply(ds, inplace=False, *, njobs=1)[source]

Apply the filter to the input dataset.

Parameters

  • ds (xarray.Dataset) – The input dataset

  • inplace (bool, optional) – If True, overwrite the input data inplace (default: False).

  • njobs (int, optional) – Number of jobs to run in parallel. Setting njobs to -1 uses the number of available cores. Disable parallelism by setting njobs to 1 (default).

Returns

The filtered dataset

Return type

xarray.Dataset

dims = ()
kwargs = {}
per_variable = True
supports_complex = True
class nd.filters.Filter(*args, **kwargs)[source]

Bases: nd.algorithm.Algorithm

The base class for a generic filter.

Parameters

dims (tuple of str) – The dimensions along which the filter is applied.

apply(ds, inplace=False, *, njobs=1)[source]

Apply the filter to the input dataset.

Parameters

  • ds (xarray.Dataset) – The input dataset

  • inplace (bool, optional) – If True, overwrite the input data inplace (default: False).

  • njobs (int, optional) – Number of jobs to run in parallel. Setting njobs to -1 uses the number of available cores. Disable parallelism by setting njobs to 1 (default).

Returns

The filtered dataset

Return type

xarray.Dataset

dims = ()
per_variable = True
supports_complex = False
class nd.filters.GaussianFilter(dims=('y', 'x'), sigma=1, **kwargs)[source]

Bases: nd.filters.Filter

A Gaussian filter.

Parameters

  • dims (tuple of str, optional) – The dimensions along which to apply the Gaussian filtering (default: (‘y’, ‘x’)).

  • sigma (float or sequence of float) – The standard deviation for the Gaussian kernel. If sequence, this is set individually for each dimension.

  • kwargs (dict, optional) – Extra keyword arguments passed on to scipy.ndimage.filters.gaussian_filter.

Returns

The filtered dataset.

Return type

xarray.Dataset

apply(ds, inplace=False, *, njobs=1)[source]

Apply the filter to the input dataset.

Parameters

  • ds (xarray.Dataset) – The input dataset

  • inplace (bool, optional) – If True, overwrite the input data inplace (default: False).

  • njobs (int, optional) – Number of jobs to run in parallel. Setting njobs to -1 uses the number of available cores. Disable parallelism by setting njobs to 1 (default).

Returns

The filtered dataset

Return type

xarray.Dataset

dims = ()
per_variable = True
supports_complex = False
class nd.filters.NLMeansFilter(dims=('y', 'x'), r=1, sigma=1, h=1, f=1, n_eff=- 1)[source]

Bases: nd.filters.Filter

Non-Local Means (Buades2011).

Buades, A., Coll, B., & Morel, J.-M. (2011). Non-Local Means Denoising. Image Processing On Line, 1, 208–212. https://doi.org/10.5201/ipol.2011.bcm_nlm

Parameters

  • dims (tuple of str) – The dataset dimensions along which to filter.

  • r ({int, sequence}) – The radius

  • sigma (float) – The standard deviation of the noise present in the data.

  • h (float) –

  • f (int) –

  • n_eff (float, optional) – The desired effective sample size. If given, must be greater than 1 and should be no larger than about half the pixels in the window. -1 means no fixed effective sample size (default: -1).

apply(ds, inplace=False, *, njobs=1)[source]

Apply the filter to the input dataset.

Parameters

  • ds (xarray.Dataset) – The input dataset

  • inplace (bool, optional) – If True, overwrite the input data inplace (default: False).

  • njobs (int, optional) – Number of jobs to run in parallel. Setting njobs to -1 uses the number of available cores. Disable parallelism by setting njobs to 1 (default).

Returns

The filtered dataset

Return type

xarray.Dataset

dims = ()
per_variable = False
supports_complex = False
nd.filters.boxcar(ds, inplace=False, dims=('y', 'x'), w=3, *, njobs=1, **kwargs)[source]

Wrapper for nd.filters.BoxcarFilter.

A boxcar filter.

Parameters

  • ds (xarray.Dataset) – The input dataset

  • inplace (bool, optional) – If True, overwrite the input data inplace (default: False).

  • dims (tuple of str, optional) – The dimensions along which to apply the filter (default: (‘y’, ‘x’)).

  • w (int) – The width of the boxcar window. Should be an odd integer in order to ensure symmetry.

  • njobs (int, optional) – Number of jobs to run in parallel. Setting njobs to -1 uses the number of available cores. Disable parallelism by setting njobs to 1 (default).

  • kwargs (dict, optional) – Extra keyword arguments passed on to scipy.ndimage.filters.convolve.

Returns

The filtered dataset

Return type

xarray.Dataset

nd.filters.convolution(ds, inplace=False, dims=('y', 'x'), kernel=None, *, njobs=1, **kwargs)[source]

Wrapper for nd.filters.ConvolutionFilter.

Kernel-convolution of an xarray.Dataset.

Parameters

  • ds (xarray.Dataset) – The input dataset

  • inplace (bool, optional) – If True, overwrite the input data inplace (default: False).

  • dims (tuple, optional) – The dataset dimensions corresponding to the kernel axes (default: (‘y’, ‘x’)). The length of the tuple must match the number of dimensions of the kernel.

  • kernel (ndarray) – The convolution kernel.

  • njobs (int, optional) – Number of jobs to run in parallel. Setting njobs to -1 uses the number of available cores. Disable parallelism by setting njobs to 1 (default).

  • kwargs (dict, optional) – Extra keyword arguments passed on to scipy.ndimage.filters.convolve.

Returns

The filtered dataset

Return type

xarray.Dataset

nd.filters.gaussian(ds, inplace=False, dims=('y', 'x'), sigma=1, *, njobs=1, **kwargs)[source]

Wrapper for nd.filters.GaussianFilter.

A Gaussian filter.

Parameters

  • ds (xarray.Dataset) – The input dataset

  • inplace (bool, optional) – If True, overwrite the input data inplace (default: False).

  • dims (tuple of str, optional) – The dimensions along which to apply the Gaussian filtering (default: (‘y’, ‘x’)).

  • sigma (float or sequence of float) – The standard deviation for the Gaussian kernel. If sequence, this is set individually for each dimension.

  • njobs (int, optional) – Number of jobs to run in parallel. Setting njobs to -1 uses the number of available cores. Disable parallelism by setting njobs to 1 (default).

  • kwargs (dict, optional) – Extra keyword arguments passed on to scipy.ndimage.filters.gaussian_filter.

Returns

The filtered dataset

Return type

xarray.Dataset

nd.filters.nlmeans(ds, inplace=False, dims=('y', 'x'), r=1, sigma=1, h=1, f=1, n_eff=- 1, *, njobs=1)[source]

Wrapper for nd.filters.NLMeansFilter.

Non-Local Means (Buades2011).

Buades, A., Coll, B., & Morel, J.-M. (2011). Non-Local Means Denoising. Image Processing On Line, 1, 208–212. https://doi.org/10.5201/ipol.2011.bcm_nlm

Parameters

  • ds (xarray.Dataset) – The input dataset

  • inplace (bool, optional) – If True, overwrite the input data inplace (default: False).

  • dims (tuple of str) – The dataset dimensions along which to filter.

  • r ({int, sequence}) – The radius

  • sigma (float) – The standard deviation of the noise present in the data.

  • h (float) –

  • f (int) –

  • n_eff (float, optional) – The desired effective sample size. If given, must be greater than 1 and should be no larger than about half the pixels in the window. -1 means no fixed effective sample size (default: -1).

  • njobs (int, optional) – Number of jobs to run in parallel. Setting njobs to -1 uses the number of available cores. Disable parallelism by setting njobs to 1 (default).

Returns

The filtered dataset

Return type

xarray.Dataset