---

name: transform-registration description: "Geometric transforms, warps, pyramids, Radon/Hough helpers, phase cross-correlation, and optical flow for scikit-image." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Transform and Registration

Use this sub-skill when you need to align images, warp coordinates, change scale, build pyramids, reconstruct from projections, or estimate dense motion with scikit-image. Keep point-feature extraction, descriptor matching, and segmentation in other routes.

Route by Task

• Warp, rotate, rescale, resize, swirl, or polar-warp an image: use references/workflows.md#warp-resize-and-pyramids.
• Apply a local nonlinear warp or custom coordinate field: use references/workflows.md#custom-coordinate-fields-and-swirl.
• Estimate a transform from point pairs: use references/workflows.md#estimate-a-transform-from-point-pairs.
• Run Radon or Hough geometry helpers: use references/workflows.md#radon-and-hough-workflows.
• Register translation, rotation, or scale: use references/workflows.md#phase-cross-correlation-and-log-polar-registration.
• Estimate dense motion fields: use references/workflows.md#optical-flow-registration.

Core Rules

• warp takes an inverse map. Pass a transform object, its .inverse, a 3×3 matrix, a callable inverse map, or a coordinate array. For 2-D homographies, SimilarityTransform, AffineTransform, and ProjectiveTransform use a faster matrix path.
• resize, rescale, resize_local_mean, warp_polar, and pyramid_* accept channel_axis; warp and rotate assume a normal image layout, so move channels before calling them if needed.
• Use preserve_range=True when the input intensities must stay on their native scale. Interpolated warps usually return floating output; use order=0 for discrete masks or edge maps.
• phase_cross_correlation returns the correction shift in input axis order, not the original motion. If the images may differ by more than half the frame, use disambiguate=True.
• optical_flow_ilk and optical_flow_tvl1 require grayscale images and return one flow component per axis.
• radon and iradon use projection angles in degrees. hough_line and probabilistic_hough_line use theta in radians.
• If from_estimate or estimate_transform fails, check bool(tform) before warping.
• Use matrix_transform for point arrays and warp_coords for reusable custom coordinate grids.

What Stays Out

• If you need corners, descriptors, or matched keypoints first, use analysis.
• If you need thresholds, morphology, masks, or region cleanup first, use segmentation-and-shapes.
• If you only need intensity normalization or denoising, use enhancement.

Reference Files

• references/workflows.md for the main geometry, registration, tomography, and motion recipes.
• references/troubleshooting.md for inverse-map, axis-order, dtype, mask, and angle-unit fixes.