Module convolution3d 

3D Convolution and Volumetric Filtering

This module provides 3D convolution and cross-correlation, separable Gaussian filtering, box filtering, non-separable median filtering, and Laplacian-of-Gaussian (LoG) blob detection for volumetric (Array3<f64>) data.

Border / Padding Modes

Three padding strategies are available via Padding3D:

Variant	Behaviour at the boundary
Zero	Out-of-bounds indices map to 0.0
Reflect	Mirror reflection around the edge voxel
Replicate	Clamp-to-edge (nearest-neighbour extrapolation)

Separable Filter Optimization

Filters that are separable (Gaussian, uniform/box) are implemented as three successive 1D convolutions along z, y, and x. This reduces the per-voxel work from O(k³) to O(3k) where k is the 1D kernel half-width.

Enums

Padding3D

Padding strategy for 3D convolution / filtering operations.

Functions

convolve3d

3D discrete convolution.

correlate3d

3D cross-correlation.

gaussian_filter3d

3D separable Gaussian filter with per-axis sigma values.

laplacian_of_gaussian3d

3D Laplacian of Gaussian (LoG) filter for blob detection.

median_filter3d

3D median filter with a rectangular window.

uniform_filter3d

3D separable uniform (box) filter with per-axis window sizes.