mincnlm

Non-local means denoising filter for MINC volumes.

mincnlm [options] <infile.mnc> <outfile.mnc>

DESCRIPTION

mincnlm applies a non-local means (NLM) denoising filter to a MINC volume. The implementation is by Pierrick Coupe and supports multiple noise models including Gaussian, Rician, and Speckle. It can operate in a voxelwise or block-based mode for improved performance.

The filter computes weighted averages of similar patches within a search neighborhood, where similarity is measured according to the selected weighting function.

OPTIONS

-sigma value

Set the noise standard deviation. Default: 0 (auto-estimate).

-beta value

Set the smoothing parameter that controls filter strength. Default: 1.

-v n

Set the neighborhood connectivity: 1 for 26-connected, 2 for 124-connected. Default: 1.

-d n

Set the search volume half-size in voxels. Default: 5.

-w n

Set the weighting function: 0 for L2 Gaussian, 1 for Pearson Speckle, 2 for L2 with Rician bias correction. Default: 0.

-aniso

Enable anisotropic filtering.

-block n

Enable block-based approach (1) or disable it (0). Default: 1.

-b_space n

Set the block spacing distance. Default: 2.

-m_min value

Set the minimum mean ratio for block matching. Default: 0.95.

-v_min value

Set the minimum variance ratio for block matching. Default: 0.5.

-mt n

Set the number of threads for parallel processing. Default: 4.

-hallucinate

Enable hallucination mode for reconstructing missing data.

-verbose

Print progress information during processing.

-debug

Enable debug output.

-clobber

Overwrite existing output files.

EXAMPLES

Denoise a volume with default settings:

mincnlm input.mnc output.mnc

Denoise with Rician noise model and known noise level:

mincnlm -w 2 -sigma 15 input.mnc output.mnc

Denoise using 8 threads with increased search radius:

mincnlm -mt 8 -d 7 input.mnc output.mnc

AUTHOR

Pierrick Coupe - McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University.

COPYRIGHTS

Copyright © 2011 by Pierrick Coupe

SEE ALSO

minc_anlm, noise_estimate