Please include necessary information to complete each entry. The
release manager should be able to step through this file without
external references.

Version 0.28
------------
* Complete deprecation of `skimage.morphology.binary_*` functions.
  - Remove reference to `skimage.morphology.binary_erosion` in docstring for `skimage.morphology.erosion`.
  - Remove reference to `skimage.morphology.binary_dilation` in docstring for `skimage.morphology.dilation`.
* Remove `skimage/graph/_mcp.pyx::MCP::find_costs` parameter `max_cumulative_cost` and `max_cost`
* Remove all traces of deprecated parameter `num_threads` in
  `skimage.restoration.rolling_ball`.
* Remove all traces of deprecated parameter `num_workers` in
  `skimage.restoration.cycle_spin`.

Version 0.27
------------
* Remove deprecated `skimage.io.imshow`,
  `skimage.io.imshow_collection`, `skimage.io.show`. Remove related plugins.
* Complete deprecation of plugin infrastructure by doing the following:
  - Replace calls to `call_plugin` with functions from the imageio plugin
  - Remove `skimage/io/_plugins`
  - Remove `skimage/io/manage_plugins.py`
  - Remove `skimage/io/__init__.py::__getattr__`
  - Remove related tests, imports, etc.
  - Remove `doc/source/user_guide/plugins.rst`
* Complete deprecation of `square`, `rectangle` and `cube` in `skimage.morphology`.

Other
-----
* Once NumPy 1.25.0 is minimal required version:
    * Remove optional test for a NaN-related deprecation warning from numpy.clip in
      skimage/exposure/tests/test_exposure.py::test_rescale_nan_warning
* Delete FREE_THREADED_BUILD block in `.github/scripts/setup-*-env.sh` once free-threaded
  wheels of Cython are available on PyPi.
* Update `central_pixel` once `sp.sparse.csgraph` supports the array interface (scipy >= 1.15)
* Finalize ``skimage.future.manual_segmentation`` API,
  see https://github.com/scikit-image/scikit-image/issues/2624
* Review why test precision needs to be lowered for macOS ARM in
  `test_apply_parallel.py`
* Once SciPy 1.17.0 is minimal required version, so that the `rng` parameter of
  `scipy.linalg.eigsh` is available everywhere, remove the compatibility logic in
  `skimage/graph/_graph_cut.py::_ncut_relabel` and the `xfail` mark in
  `tests/skimage/graph/test_rag.py::test_reproducibility`.

Post numpy 2
------------
- Remove try except blocks following comment starting
  `# TODO: when minimum numpy`
  in `skimage/color/colorconv.py`, `skimage/color/tests/test_colorconv.py`,
  `skimage/measure/_blur_effect.py`, and `skimage/util/tests/test_montage.py`
- Remove `handle_np2` from `skimage/conftest.py`
- Once matplotlib doesn't constrain to numpy<2.0 via contourpy (or 2.0 is
  minimal required version): consider re-adding matplotlib to the test
  dependencies. Also remove temporary assert for NumPy 2.0 in
  `nightly_wheel_build.yml`.
- Remove references to `numpy.bool8` once NumPy 2.0 is minimal required version.
- scipy>=1.12: remove SCIPY_CG_TOL_PARAM_NAME in `_shared.compat.py`.
- Remove `np2` check in `skimage/feature/brief.py`.

Version 2.0
-----------
- In `skimage.morphology.remove_small_objects`, complete the deprecation of
  `min_size` and re-order the argument `connectivity` after the
  keyword-only parameter `max_size`.
- In `skimage.morphology.remove_small_holes`, complete the deprecation of
  `area_threshold` and re-order the argument `connectivity` after the
  keyword-only parameter `max_size`.
- In `skimage.measure.regionprops` complete deprecation of old property names.
  Remove `PROPS` dict, related logic and warnings.

See https://github.com/scikit-image/scikit-image/wiki/API-changes-for-skimage2

Version 2.2
-----------
- In `skimage.measure.fit` remove the deprecated `estimate` method and simplify
  the internals that where providing backwards-compatibility with this API.
  This affects the classes `LineModelND`, `CircleModel`, and `EllipseModel`.
- In `skimage.measure._geometric` and `skimage.measure._thin_plate_splines`,
  remove the deprecated `estimate` method and simplify the internals that where
  providing backwards-compatibility with this API.
  This affects the classes `AffineTransform`, `EssentialMatrixTransform`,
  `EuclideanTransform`, `FundamentalMatrixTransform`,
  `PiecewiseAffineTransform`, `PolynomialTransform`, `ProjectiveTransform`,
  `SimilarityTransform`, and `ThinPlateSplineTransform`

- Complete deprecation of:
  * `params` argument to `predict_*` methods of `LineModelND`, `CircleModel`
    and `EllipseModel` from `skimage.measure.fit`.
  * Ability to create uninitialized models of the same classes, by passing no
    input arguments to constructor.

  The procedure is:

  * Move `_estimate` code into `from_estimate` (this as part of expiring
    deprecation of `estimate` methods).
  * Delete deprecated `BaseModel`, and `_BaseModel` class for supporting deprecations.
  * Delete `_deprecate_no_args` (class decorator to deprecate no-argument call
    to class constructor).
  * Delete `_deprecate_model_params` (method decorator to deprecate `params`
    argument to `predict_*` methods).
  * Remove `params` arguments (and parsing) to `predict_*` methods.
  * Delete `_params2init_values` methods (these were for processing deprecated `params` arguments).
  * Delete `params` properties of models.
  * Move `_check_init_values` into `_args_init` (these were for checking *both*
    deprecated `params` arguments, *and* `__init__` arguments; after expiry,
    the code only needs to check `__init__` arguments).
  * Rename `_arg_init` methods to `__init__`.
  * Remove tests for `params` arguments, and no-args initialization.  These
    will become apparent when running the test suite after making the changes above.
