NormalizeStandard

class deeptrack.optical.math.NormalizeStandard(featurewise: bool | Callable[[...], bool] = True, channel_axis: int | None = -1, **kwargs: Any)

Bases: Feature

Standardize an array to zero mean and unit variance.

Applies z-score normalization:

output = (input - mean) / std

where the standard deviation is computed as the population standard deviation (dividing by N, not N-1).

Axis semantics: - If featurewise=False, normalization is applied globally. - If featurewise=True and channel_axis is specified, normalization is

applied independently per channel.

• If featurewise=True and channel_axis=None, normalization defaults to global behavior.

The output always preserves the input shape.

Parameters

featurewise: bool, optional

Whether to normalize each feature independently. It default to True, which is the only behavior currently implemented.

Returns

np.ndarray or torch.Tensor

Standardized array with the same shape as input.

Methods

get(image: array, **kwargs: Any) -> array

Standardizes the input image to mean 0 and std deviation 1.

Examples

>>> import deeptrack as dt

Create an input image.

>>> import numpy as np
>>>
>>> input_image = np.array([[1, 2], [3, 4]], dtype=float)

>>> standardizer = dt.NormalizeStandard()
>>> output_image = standardizer(input_image)
>>> output_image
array([[-1.34164079, -0.4472136 ],
    [ 0.4472136 ,  1.34164079]])

Methods Summary

get(image, featurewise[, channel_axis])

Standardize the input image to zero mean and unit variance.

Methods Documentation

get(image: ndarray | Tensor, featurewise: bool, channel_axis: int | None = -1, **kwargs: Any) → ndarray | Tensor

Standardize the input image to zero mean and unit variance.

Applies z-score normalization:

(image - mean) / std

where std is the population standard deviation (i.e., computed with denominator N).

Axis semantics: - If featurewise=False, normalization is applied globally over all elements. - If featurewise=True and channel_axis is specified, normalization is applied independently along each channel. - If featurewise=True and channel_axis=None, normalization falls back to global behavior.

The output preserves the input shape.

Parameters

image: np.ndarray or torch.Tensor

Input array to standardize. Must match the selected backend.

featurewise: bool

Whether to normalize each channel independently.

channel_axis: int or None, optional

Axis corresponding to channels/features. If None, no channel-wise normalization is performed.

**kwargs: Any

Additional keyword arguments (unused).

Returns

np.ndarray or torch.Tensor

Standardized array with the same shape and backend as the input.