pad_image_to_fft

deeptrack.optical.math.pad_image_to_fft(image: ndarray | Tensor, axes: Iterable[int] = (0, 1)) → ndarray | Tensor

Pad an image to improve Fast Fourier Transform (FFT) performance. Padding is applied at the end of each axis (no centering).

Preserves backend: - NumPy input → NumPy output - Torch input → Torch output (preserves autograd compatibility)

This function pads an image by adding zeros to the end of specified axes so that their lengths match the nearest larger size in _FASTEST_SIZES. Sizes are chosen as products of small prime factors, which are efficient for FFT algorithms.

Parameters

image: np.ndarray | torch.Tensor

The input image to pad.

axesiterable of int, optional

Axes along which to apply padding. Negative axes are supported.

Returns

np.ndarray | torch.Tensor

The padded image with dimensions optimized for FFT performance.

Raises

ValueError

If no suitable size is found in _FASTEST_SIZES for any axis length.

Examples

>>> import numpy as np
>>> from deeptrack.image import pad_image_to_fft

Pad a NumPy array:

>>> img = np.zeros((5, 11))
>>> padded_img = pad_image_to_fft(img)
>>> print(padded_img.shape)
(6, 12)