mmcfilters

mmcfilters is an alpha research library for constructing and editing morphological hierarchies, computing node attributes, and applying connected filters and hierarchy operators. It provides a C++20 core and Python bindings.

The library represents hierarchies through a rooted connected-subset tree model. A morphological tree of partial partitions is the stricter case in which every node has a non-empty proper part. All supported image-domain trees use the same topology model, with an explicit smallest-node map and optional regular-grid geometry and adjacency.

Use mmcfilters for experiments that require mutable tree topology, direct ownership of hierarchy state, incremental attributes, residual hierarchies, or the filtering operators implemented by this project. For stable, general-purpose hierarchical image processing, Higra will often be the better choice.

Main capabilities

• construction of max-trees, min-trees, trees of shapes, and self-dual residual trees;
• generic rooted hierarchies with safe local and staged edits;
• a broad catalog of node attributes organized into gray-level, shape, moment, boundary, and tree-topology groups;
• a range of attribute-based filters and hierarchy operators, including pruning and non-pruning strategies;
• pixel contours and ordered geometric contour traces;
• import, export, and attribute projection for Higra-style hierarchies;
• typed altitude handling in C++ and a focused np.uint8 Python API.

Requirements

• Python 3.9–3.14 for the distributed Python package;
• NumPy 1.23 or newer;
• a C++20 compiler and CMake 3.20 or newer for source builds.

Installation

Install the Python package from PyPI:

python -m pip install mmcfilters

Install from a source checkout:

python -m pip install .

For a C++-only build and installation:

cmake -S . -B build -DMMCFILTERS_BUILD_PYTHON=OFF
cmake --build build
cmake --install build --prefix /path/to/prefix

The installed C++ package exports the mmcfilters::core target:

find_package(mmcfilters CONFIG REQUIRED)
target_link_libraries(my_target PRIVATE mmcfilters::core)

Quick start

import numpy as np
import mmcfilters

image = np.ascontiguousarray(
    [
        [3, 3, 2, 2],
        [3, 4, 4, 2],
        [1, 4, 5, 2],
        [1, 1, 5, 0],
    ],
    dtype=np.uint8,
)

# radius=1.5 selects 8-connectivity; use 1.0 for 4-connectivity.
tree = mmcfilters.MorphologicalTreeFactory.create_max_tree(image, radius=1.5)

root = tree.root
root_children = tree.children(root)

pixel = 10
smallest = tree.smallest_node(pixel)
support_pixels = list(tree.node_support(smallest))
component_mask = tree.reconstruct_node(smallest)

area = mmcfilters.Attribute.compute_single_topology_attribute(
    tree,
    mmcfilters.Attribute.AREA,
)
max_dist = mmcfilters.Attribute.compute_single_attribute(
    tree,
    mmcfilters.Attribute.MAX_DIST,
)
reconstructed_image = tree.reconstruct_from_node_altitudes()

Documentation

User guides

Reproducibility

• SIBGRAPI 2026 — Unifying Local Attribute Computation on Component Trees and the Tree of Shapes

References

• Attribute catalog
• Generated C++ API

Contributor guides

• Attribute computer architecture
• Finite-window local-attribute C++ extension
• Scientific benchmark builds

License

mmcfilters is distributed under the GNU General Public License v3.0.