decompose op• Datenarten: image2d → table
• Aufruf: import motionmag; motionmag.complex_steerable_decompose(image, scales: 'int' = 4, orientations: 'int' = 4) -> 'dict' (oder opsmotionmag.get("complex_steerable_decompose"))
Komplexe orientierte Teilbandzerlegung eines Bildes -> `dict`.
> Die ausführliche Beschreibung unten ist der Originaltext — Zusammenfassung und Überschriften sind übersetzt.
Splits the image into `scales * orientations` analytic sub-bands plus
three residuals (low-pass, high-pass and a small symmetric completion band).
Each sub-band is a full-resolution `(H, W)` complex array whose modulus is
the local contrast of that scale/orientation and whose argument is the local
phase — the quantity a translation shifts linearly, which is what the
rest of this module is built on. There is no spatial decimation: keeping
every band at full resolution costs memory but makes the frame exactly
invertible, and exactness is the point.
Returns ``{"bands": [complex (H, W), ...], "kinds": [...],
"centre_cycles_per_px": [...], "orientation_rad": [...], "shape": (H, W),
"scales": s, "orientations": k}`. kinds[j] is "band"` for an oriented
sub-band and `"lowpass" / "highpass" / "residual"` otherwise;
`centre_cycles_per_px and orientation_rad are None` for residuals,
which have no orientation and no single centre frequency.
Feed the whole dict back to :func:complex_steerable_reconstruct. Round trip
error, measured on a 64x64 random frame with the defaults, is
`max|out - in| = 6.66e-16`; on a 31x37 (odd, non-square) frame 7.22e-16;
and the worst over every `scales` in 1..8 crossed with every
`orientations` in 1..16 on 32x32 is 7.77e-16 — the tight-frame construction
is exact, not approximate (see the notes on the self-conjugate grid points
in :func:_filter_bank).
References: Freeman & Adelson, IEEE PAMI 1991; Simoncelli & Freeman,
ICIP 1995; Portilla & Simoncelli, IJCV 2000.
• Leitfaden zur Familie motion_magnification
• Katalog der Beispieldaten (Download-URLs / Lizenzen) — 2-D nutzt skimage.data (BSD/Public Domain) plus synthetische Bilder, 3-D nennt Download-URLs echter Datenquellen (Stanford, PDS, …).
• Herkunft und Literatur der Operatoren — die Quellen der Forschung/Verfahren, auf denen diese Operatorfamilie beruht.
• Der kanonische Algorithmus (Autor, Jahr) und seine Anwendungen stehen im Familienleitfaden oben.
• motion_magnification — py -3.11 examples/motion_magnification.py
table als Eingabe)decompose)*Provenance: motionmag.py — MOTIONMAG Operator-Registry. Diese Notiz wird von tools/opdocs.py md erzeugt (nicht von Hand bearbeiten).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.