RicianNoise#

class deepinv.physics.RicianNoise(sigma=0.1, rng=None)[source]#

Bases: NoiseModel

RicianNoise: \(y = \sqrt{(x + \sigma \epsilon_1)^2 + (\sigma \epsilon_2)^2}\)

where \(\epsilon_1\sim\mathcal{N}(0,I)\) and \(\epsilon_2\sim\mathcal{N}(0,I)\)

This noise model is often used in MRI imaging and has the property of keeping pixel intensities \(\geq 0\)

Warning

All pixel intensities will become positive: this noise model may not be suited for data with negative intensities.

Parameters:

  • sigma (Union[float, torch.Tensor]) – Standard deviation used.

  • rng (torch.Generator, None) – (optional) a pseudorandom random number generator for the parameter generation.

forward(x, sigma=None, seed=None)[source]#

Adds the noise to measurements x

Parameters:

  • x (torch.Tensor) – measurements

  • sigma (float, torch.Tensor, None) – standard deviation to be used. If not None, it will overwrite the current noise level.

  • seed (int, None) – the seed for the random number generator.

Returns:

noisy measurements