phase_correlation_shift — FIT registration op

• Data kinds: image2d × image2d → table

• Call: import fullseye as fs; fs.ledger.phase_correlation_shift(image1, image2, window: 'str | None' = 'hann', whitening: 'float' = 0.5, method: 'str' = 'upsample', upsample_factor: 'int' = 100) -> 'dict' (to call the implementation directly, import filters_freq; filters_freq.phase_correlation_shift(image1, image2, window: 'str | None' = 'hann', whitening: 'float' = 0.5, method: 'str' = 'upsample', upsample_factor: 'int' = 100) -> 'dict'; from the registry, opsfit.get("phase_correlation_shift"))

• GPU: none (CPU reference implementation only)

Usage

Estimate the subpixel translation needed to align two images by phase-only correlation.

The default method="upsample" refines the whitening-aware periodic FFT peak by evaluating a 1.5-pixel neighbourhood with a matrix-multiply local DFT using only NumPy (upsample_factor=100 by default). method="parabola" is faster, but its wrapped three-point parabola is biased about 0.166 px toward the integer grid. On the fixed 96x112 truth gates, parabola errors are 0.166 px (Hann) / 0.172 px (no window), versus 0.010 px for upsampling with either window; upsampling is required to stay within 0.02 px and within 0.01 px of scikit-image when it is installed. The local DFT follows Guizar-Sicairos, Thurman & Fienup (2008), Optics Letters 33(2), 156-158, doi:10.1364/OL.33.000156. The sign convention is explicit: when image2 = shift(image1, d), the result is approximately -d, so shift(image2, result) approximately equals image1. Returns row_shift/col_shift (also dy/dx), peak and peak_ratio (largest peak divided by the best peak outside its 3x3 neighbourhood). Constant or relatively textureless inputs raise ValueError.

References (sample data, literature)

• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).

• Operator provenance and references — the sources of the research/methods this op family came from.

• The canonical algorithm (author, year) and its uses are named in the family usage guide above.

Runnable examples (verified samples that actually call this op)

• fit_ops_from_knowhow — py -3.11 examples/fit_ops_from_knowhow.py

Ops the type connects to (they accept table as input)

—

Same category (registration)

—

*Provenance: filters_freq.py — FIT operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*

© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.