clearex.registration.tre

Functions

brief_descriptors_at_keypoints(img, keypoints_rc)

Compute BRIEF descriptors at a set of row/column keypoints.

cv_tre_proxy(src_xy, dst_xy[, n_splits, ...])

Estimate a proxy target-registration error by repeated holdout.

fit_transform_ransac(src_xy, dst_xy[, ...])

Fit a robust geometric transform using RANSAC.

mutual_nn_pairs(fixed_rc, moving_rc[, max_dist])

Match points by mutual nearest-neighbor assignment.

particle_registration_tre(img1, img2[, ...])

Register two particle images and summarize landmark error metrics.

registration_errors(model, src_xy, dst_xy)

Compute residual summary statistics for a fitted transform.

xy_from_rc(rc)

Convert row/column coordinates to x/y coordinates.
clearex.registration.tre.brief_descriptors_at_keypoints(img, keypoints_rc, patch_size=31, n_bits=256)

Compute BRIEF descriptors at a set of row/column keypoints.

Parameters:

  • img (numpy.ndarray) – Input 2D image.

  • keypoints_rc (numpy.ndarray) – Candidate keypoints in (row, col) format.

  • patch_size (int, default=31) – BRIEF patch size in pixels.

  • n_bits (int, default=256) – Descriptor length in bits.

Returns:

Rounded keypoints that produced valid descriptors and their BRIEF descriptor matrix.

Return type:

tuple of numpy.ndarray

clearex.registration.tre.xy_from_rc(rc)

Convert row/column coordinates to x/y coordinates.

Parameters:

rc (numpy.ndarray) – Array of coordinates in (row, col) order.

Returns:

Coordinates in (x, y) order.

Return type:

numpy.ndarray

clearex.registration.tre.fit_transform_ransac(src_xy, dst_xy, model='similarity', residual_threshold=2.0, max_trials=1000)

Fit a robust geometric transform using RANSAC.

Parameters:

  • src_xy (numpy.ndarray) – Source points in (x, y) order.

  • dst_xy (numpy.ndarray) – Destination points in (x, y) order.

  • model ({"similarity", "affine"}, default="similarity") – Transform family used by RANSAC.

  • residual_threshold (float, default=2.0) – Maximum inlier residual in pixels.

  • max_trials (int, default=1000) – Maximum number of RANSAC iterations.

Returns:

Robust transform model and the Boolean inlier mask returned by skimage.measure.ransac().

Return type:

tuple

clearex.registration.tre.registration_errors(model, src_xy, dst_xy)

Compute residual summary statistics for a fitted transform.

Parameters:

  • model (skimage.transform.SimilarityTransform or skimage.transform.AffineTransform) – Transform used to map source points into destination space.

  • src_xy (numpy.ndarray) – Source points in (x, y) order.

  • dst_xy (numpy.ndarray) – Target points in (x, y) order.

Returns:

Residual vector, RMS residual, median residual, and 95th percentile residual, all in pixels.

Return type:

tuple

clearex.registration.tre.cv_tre_proxy(src_xy, dst_xy, n_splits=20, test_fraction=0.2, random_state=0, model='similarity')

Estimate a proxy target-registration error by repeated holdout.

Parameters:

  • src_xy (numpy.ndarray) – Source points in (x, y) order.

  • dst_xy (numpy.ndarray) – Destination points in (x, y) order.

  • n_splits (int, default=20) – Number of random holdout repetitions.

  • test_fraction (float, default=0.2) – Fraction of matches assigned to each holdout split.

  • random_state (int, default=0) – Seed for the split generator.

  • model ({"similarity", "affine"}, default="similarity") – Transform family evaluated during cross-validation.

Returns:

RMS holdout error and the full array of holdout residuals.

Return type:

tuple

clearex.registration.tre.particle_registration_tre(img1, img2, fwhm_px=10.0, invert=False, thresholds=(0.03, 0.03), brief_patch_size=31, model='similarity', ransac_residual_threshold=2.0, show_debug=False, optimize=False)

Register two particle images and summarize landmark error metrics.

Parameters:

  • img1 (numpy.ndarray) – Fixed image in 2D.

  • img2 (numpy.ndarray) – Moving image in 2D.

  • fwhm_px (float, default=10.0) – Expected particle full width at half maximum in pixels.

  • invert (bool, default=False) – If True, invert both preprocessed images before blob detection.

  • thresholds (tuple of float, default=(0.03, 0.03)) – Detection thresholds for img1 and img2.

  • brief_patch_size (int, default=31) – Patch size used for BRIEF descriptors.

  • model ({"similarity", "affine"}, default="similarity") – Geometric model fitted during robust registration.

  • ransac_residual_threshold (float, default=2.0) – Inlier threshold in pixels for RANSAC.

  • show_debug (bool, default=False) – If True, plot intermediate match diagnostics.

  • optimize (bool, default=False) – If True, stop after detection and visualization to help tune detector parameters.

Returns:

Registration summary dictionary, or None when optimize=True and the helper exits after visualization.

Return type:

dict or None

Raises:

ValueError – If either input image is not 2D.

clearex.registration.tre.mutual_nn_pairs(fixed_rc, moving_rc, max_dist=None)

Match points by mutual nearest-neighbor assignment.

Parameters:

  • fixed_rc (numpy.ndarray) – Fixed points in (row, col) format.

  • moving_rc (numpy.ndarray) – Moving points in (row, col) format, typically already mapped into fixed-image space.

  • max_dist (float, optional) – Maximum distance allowed for a retained match.

Returns:

Pair indices [moving_idx, fixed_idx] and their corresponding Euclidean distances.

Return type:

tuple of numpy.ndarray