synthesizer.instruments.photometric_noise¶

Photometric noise generation for synthesizer imaging.

This module contains tools for modelling and generating photometric noise for imaging. Correlated noise is represented by the CorrelatedNoiseModel class, which stores a source noise template and derives reusable statistical quantities from it. These cached quantities are then used to generate new noise realisations for images of arbitrary shape.

Classes

class synthesizer.instruments.photometric_noise.CorrelatedNoiseModel(source_noise_map)[source]¶

Model correlated image noise from a source template.

This class encapsulates the correlated-noise machinery used by imaging. A source noise template (i.e. existing noise field) is supplied at construction time and is used to estimate a correlation function (CF). The expensive CF estimation is cached on the model so that many target images can draw fresh correlated noise realisations from the same underlying template.

source_noise_map¶

The observed/template noise map used to characterise the noise.

Type:: np.ndarray or unyt_array

units¶

Units carried by the source template, if present.

Type:: unyt.Unit or None

_cf_cache¶

Cache of estimated correlation functions keyed by (subtract_mean, correct_periodicity).

Type:: dict

apply_noise(image, correct_periodicity=True, rng_seed=None, inplace=False)[source]¶

Apply correlated noise to an Image.

The correlated-noise realisation is generated for the image’s pixel shape and then added via the image’s existing noise-array machinery.

Parameters:

image (Image) – The image to which the correlated noise should be added.
correct_periodicity (bool) – If True the DFT periodicity correction is applied.
rng_seed (int, optional) – Seed for the random number generator.
inplace (bool) – If True, update the input image in place and return it. Otherwise return a new image. Default is False.

Returns:

The noisy image.

Return type:

Image

Raises:

InconsistentArguments – If the image has units but the noise model is dimensionless.

estimate_correlation_function(subtract_mean=False, correct_periodicity=True)[source]¶

Estimate and cache the source correlation function.

Computes an unrolled correlation function (DFT convention, origin at [0, 0]) from the power spectrum of the source array. The result is cached for the requested modelling options so later calls can reuse it without recomputing the FFT.

Parameters:

subtract_mean (bool) – If True the DC component of the power spectrum is zeroed, removing the mean offset from the noise model.
correct_periodicity (bool) – If True a correction factor is applied to compensate for the periodicity assumption of the DFT.

Returns:

A 2D array of the same shape as the source template holding the unrolled CF with the origin at [0, 0].

Return type:

np.ndarray

generate_noise_array(target_shape, subtract_mean=False, correct_periodicity=True, rng_seed=None)[source]¶

Generate a correlated noise realisation for a target shape.

A cached source CF is first fetched or computed, then resampled to the target dimensions before a fresh random realisation is drawn. The same source CF can therefore be used to generate many independent target noise fields.

Parameters:

target_shape (tuple of int) – The (ny, nx) pixel dimensions of the desired noise field.
subtract_mean (bool) – Passed through to estimate_correlation_function().
correct_periodicity (bool) – Passed through to estimate_correlation_function().
rng_seed (int, optional) – Seed for the random number generator. Pass the same seed to reproduce an identical noise realisation.

Returns:

A 2D array of shape target_shape containing the generated correlated noise field. Units are attached if the source template carried them.

Return type:

np.ndarray or unyt_array

Raises:

InconsistentArguments – If the source CF is too small to contribute any pixels to the target CF grid.

property source_array: ndarray¶

Return the source noise template without units.

Returns:: The source noise template as a plain NumPy array.
Return type:: np.ndarray