itk_minc_nonlocal_filter

Patch-based non-local means denoising.

itk_minc_nonlocal_filter <input.mnc> <output.mnc> [options]

DESCRIPTION

itk_minc_nonlocal_filter performs patch-based non-local means (NLM) denoising on MINC volumes. The filter computes a weighted average of voxel neighbourhoods, where the weights are determined by the similarity between patches. This preserves edges and fine structures better than Gaussian smoothing.

Several variants are supported: the standard NLM filter, the adaptive NLM (ANLM), median-based filtering, and various output modes including similarity maps, weight maps, and standard deviation maps. The search and patch radii control the trade-off between computation time and denoising quality.

An ROI mask can restrict processing to a specific region. The noise level can be estimated automatically or specified manually.

OPTIONS

--noise f

Specify the noise standard deviation. If not given, it is estimated automatically.

--clobber

Overwrite the output file if it already exists.

--verbose

Print progress information during processing.

--search r

Search radius in voxels (default: 2).

--patch r

Patch radius in voxels (default: 1).

--float

Store output voxels as single-precision floating point.

--short

Store output voxels as short integer.

--byte

Store output voxels as unsigned byte.

--beta f

Smoothing parameter controlling the decay of weights (default: 1.0).

--mindist

Use minimum distance instead of mean distance for weight computation.

--median

Use median instead of weighted mean for output.

--similarity

Output the similarity map instead of the denoised image.

--mweight

Output the maximum weight map.

--mean

Output the local mean.

--sd

Output the local standard deviation.

--flat

Use flat (uniform) patch weighting.

--ball

Use a ball-shaped search neighbourhood instead of a cube.

--anlm

Use the adaptive non-local means variant.

--roi minc

Restrict processing to voxels within the specified ROI mask.

--log

Process the image in the log domain.

--regularize f

Apply regularization with the specified weight.

--preselect

Enable voxel preselection based on local statistics (faster).

--no_preselect

Disable voxel preselection.

--th_m f

Mean preselection threshold (default: 0.95).

--th_v f

Variance preselection threshold (default: 0.5).

EXAMPLES

# Basic non-local means denoising
itk_minc_nonlocal_filter input.mnc denoised.mnc

# ANLM with specified noise level
itk_minc_nonlocal_filter input.mnc denoised.mnc --anlm --noise 5.0

# Larger search radius with preselection
itk_minc_nonlocal_filter input.mnc denoised.mnc --search 3 --patch 2 \
    --preselect --float

# Denoise within a mask
itk_minc_nonlocal_filter input.mnc denoised.mnc --roi brain_mask.mnc

AUTHOR

Vladimir S. Fonov - McConnell Brain Imaging Centre, Montreal Neurological Institute.

COPYRIGHTS

Copyright © 2009-2024 by Vladimir S. Fonov

SEE ALSO

minc_anlm , mincnlm , noise_estimate