Parameter Reference

This page describes all parameters for the bm3d_ring_artifact_removal function.

Mode Selection

Parameter	Default	Description
`mode`	`"generic"`	Operation mode. Use `"generic"` for white noise, `"streak"` for ring artifacts.

Note

For ring artifact removal in tomography sinograms, always use mode="streak". The streak mode is specifically optimized for vertical artifacts caused by detector pixel response variations.

Noise Parameters

Parameter	Default	Description
`sigma_random`	`0.1`	Estimated noise standard deviation. Higher values = more aggressive denoising.

Guidelines for sigma_random:

Light artifacts: 0.02 - 0.05
Moderate artifacts: 0.05 - 0.10
Heavy artifacts: 0.10 - 0.20

Tip

Start with a low value and increase gradually. Check the difference image (input - output) to ensure you’re removing artifacts, not signal.

Block Matching Parameters

Parameter	Default	Description
`patch_size`	`8`	Size of patches for block matching. Use 7 or 8 for best results.
`step_size`	`4`	Stride for patch extraction. Lower = better quality but slower.
`search_window`	`24`	Maximum search distance for finding similar patches.
`max_matches`	`16`	Maximum number of similar patches per 3D group.

Quality vs Speed Tradeoffs:

Faster processing: step_size=6-8, max_matches=8
Higher quality: step_size=2-3, max_matches=32

Streak Mode Parameters

These parameters only apply when mode="streak":

Parameter	Default	Description
`streak_sigma_smooth`	`3.0`	Sigma for Gaussian smoothing in streak profile estimation.
`streak_iterations`	`2`	Number of iterations for robust streak estimation.
`sigma_map_smoothing`	`20.0`	Sigma for smoothing the spatially-varying noise map.
`streak_sigma_scale`	`1.1`	Scale factor for streak sigma estimation.
`psd_width`	`0.6`	PSD Gaussian width for streak mode filtering.

Advanced Parameters

Parameter	Default	Description
`threshold`	`2.7`	Hard thresholding coefficient for the first BM3D stage.
`batch_size`	`32`	Chunk size for 3D stack processing (controls memory usage).
`sigma_map`	`None`	Optional pre-computed sigma map for 3D processing.

Multiscale Mode (Experimental)

Warning

Multiscale mode is experimental. It works well for wide ring artifacts (>39 pixels) but may over-process regular sinograms.

Parameter	Default	Description
`multiscale`	`False`	Enable multi-scale BM3D for wide streaks. Only works with `mode="streak"`.
`num_scales`	`None`	Override automatic scale calculation. If None, uses `floor(log2(width/40))`.
`filter_strength`	`1.0`	Multiplier for BM3D filtering intensity.
`debin_iterations`	`30`	Iterations for cubic spline debinning.

Example Usage

Basic streak removal:

from bm3dornl import bm3d_ring_artifact_removal

denoised = bm3d_ring_artifact_removal(
    sinogram,
    mode="streak",
    sigma_random=0.05,
)

High quality processing:

denoised = bm3d_ring_artifact_removal(
    sinogram,
    mode="streak",
    sigma_random=0.05,
    step_size=2,
    max_matches=32,
)

Fast processing for large datasets:

denoised = bm3d_ring_artifact_removal(
    sinogram_stack,
    mode="streak",
    sigma_random=0.05,
    step_size=6,
    max_matches=8,
    batch_size=16,  # Reduce memory usage
)