Restoration

Restoration algorithms clean up a field – removing noise, resampling, smoothing – without changing what it physically represents.

Denoising

• TVD / tvd() – total-variation denoising; preserves sharp interfaces (plume edges) while removing speckle. split_bregman_tvd() and the "heterogeneous bregman" method allow a spatially varying weight.

• H1_regularization() – smoother, cheaper H1 (Tikhonov) regularization.

• Median – median filtering.

import darsia

clean = darsia.tvd(img=noisy, method="isotropic Bregman",
                   weight=0.03, max_num_iter=100)

Resampling and averaging

• Resize / resize(), equalize_voxel_size(), uniform_refinement() – change resolution.

• VolumeAveraging, REV, volume_average(), porosity_based_averaging() – upscale a fine field to representative elementary volumes.

Binary clean-up

BinaryFillHoles, BinaryRemoveSmallObjects and BinaryLocalConvexCover tidy the masks produced by segmentation and thresholding.

See also

• Total-variation and H1 regularization

• Restoration