[sdiebolt]

Hi! This lets SmoothingRecursiveGaussianImageFilter skip axes shorter than four pixels while smoothing the remaining axes, including singleton slices used in registration. Closes #6997.

Scope and local testing

• Preserve image dimensionality and geometry, and retain the existing processing order when every axis has at least four pixels.
• Cast directly to the output type when all axes are short, avoiding an unnecessary float conversion.
• Keep empty-image and nonpositive-sigma validation.
• Leave recursive Gaussian gradients for a separate issue and PR; registration still needs both metric gradient-filter flags disabled for short dimensions.

Local checks on Linux:

• Added GoogleTests for short-axis combinations, anisotropic sigma and spacing, geometry, repeated updates, in-place execution, integer/vector pixels, and two-level singleton-image registration.
• All 14 focused CTest entries passed, including the unchanged smoothing MD5 baseline; the image-adaptor test also passed when invoked directly.
• A new regression test compiled against unmodified main headers failed with the expected four-pixel exception.
• Rebuilt the Python registration bindings and ran the exact example from ENH: Support singleton and short dimensions in SmoothingRecursiveGaussianImageFilter #6997 unchanged. Both sigma=1.0 and sigma=0.0 recovered the expected 4-pixel X translation.
• All local commit hooks passed. Full-suite, cross-platform, and performance validation remain pending.

PR Checklist

PR Checklist

• No API changes were made (or the changes have been approved)
• No major design changes were made (or the changes have been approved)
• Added test (or behavior not changed)
• Updated API documentation (or API not changed)
• Added license to new files (if any)
• Added Python wrapping to new files (if any) as described in ITK Software Guide Section 9.5
• Added ITK examples for all new major features (if any)

No public API additions or new wrapped classes. The class documentation describes the short-axis behavior; both new test files have the standard license header. No separate examples contribution is needed for this enhancement.

AI assistance

AI assistance contributed to diagnosis, implementation, and regression tests. Focused C++ and Python checks passed locally.

[sdiebolt]

The Linux Python failure looks unrelated to this change: Gaussian smoothing passes, and the only failure among 185 tests is the median filter’s type-hint check (NameError: name 'xr' is not defined). The affected type alias and test are unchanged by this PR. This matches the failure observed in #7002.

[dzenanz]

Mostly looks good. I have not taken such a good look at tests besides ShortAxesAndRepeatedUpdates.

[dzenanz]

This largely duplicates filter's functionality. Whenever the filter is updated, this needs to be updated too. I would prefer a simpler test, with some simple-to-reason-about constants. If we want to make sure that the filter's output does not drift, we should add a new image regression baseline with the expected output. That is better/easier to do via classic test, as opposed to GTest.

[hjmjohnson]

Rebase only: onto current upstream/main. The conflict was in RegistrationMethodsv4/test/CMakeLists.txt; the new ShortDimensions GTest now sits in the same list as the SyN GTest, with one driver call. Build and the short-axis and Smoothing GTests pass locally.