typed op• Datenarten: counts → counts
• Aufruf: fullseye.apply(img, "tb_tcspc_background_subtract", a=0.5, b=0.5) (das 2-D-Modell ist ein Bild plus zwei skalare Regler a,b∈[0,1])
*Keine Abbildung: dieser Op nimmt counts als Eingabe. Ein Studio-Programm, das bei einem Bild beginnt, erreicht diesen Typ nicht — siehe die lauffähigen Beispiele unten.*
Entfernen des Umgebungslicht-/Dunkelzähl-Untergrunds aus einem Ankunftszeit-Histogramm.
> Die ausführliche Beschreibung unten ist der Originaltext — Zusammenfassung und Überschriften sind übersetzt.
Outdoors, most of what a dToF sensor counts is sunlight: a roughly uniform
pedestal under the return pulse. It biases the centroid toward the middle of
the window (a floor of `b` per bin pulls the first moment toward
`window/2`) and it inflates the apparent signal, so it is removed before
any depth or lifetime estimate.
The level is estimated by *method* and then subtracted (the sign trap:
the result is `hist - level, clipped at 0, never hist + level`):
• `"median"` (default) — the median of every bin. Robust while the pulse
occupies well under half the window, which is the normal dToF case.
• `"leading"` — the mean of the first *leading_bins* bins, the classical
choice when the pulse is known to arrive late (a far target).
• `"trailing"` — the mean of the last *leading_bins* bins, for
fluorescence decays where the tail is background.
• `"quantile"` — the given *quantile* of all bins, for tuning by hand.
*leading_bins* defaults to `None = min(8, len(hist))`, so the default
call works on a short histogram instead of raising over a constant nobody
chose (a fixed default of 8 made `method="leading"` fail on any histogram
with fewer than 8 bins).
*scale* multiplies the estimated level before subtraction (`scale=1.2` for
a deliberately aggressive removal). Clipping at 0 means the result is a valid
non-negative histogram that the rest of this module will accept.
Ground truth: on a noiseless histogram with a known flat pedestal of 20
counts/bin under a 5000-photon pulse covering 5.1% of the window, the median
estimate recovers 20.000000 and the returned histogram equals the pedestal-
free pulse exactly (measured area error 0.0, pinned in the tests).
Returns a float64 1-D histogram of the same length as *hist*.
Raises `ValueError`: negative, non-finite or non-1-D *hist*, an unknown
*method*, a *leading_bins* outside `[1, len(hist)]`, a *quantile* outside
`[0, 1]`, and a negative *scale*.
Typed bridge of the photon op `tcspc_background_subtract into the 2-D evolution registry: the same implementation, called under the op(v, a, b) convention. a drives quantile (default 0.5) and b drives scale` (default 1).
• 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.
• (noch keine)
counts als Eingabe)identity · tb_spad_deadtime_apply · tb_spad_deadtime_correct · tb_tcspc_coates_correct · tb_tcspc_irf_convolve · tb_dtof_depth · tb_countrate_to_counts · tb_counts_to_countrate
typed)tb_points_to_voxel · tb_estimate_point_normals · tb_iss_keypoints · tb_angle_3points · tb_project_points · tb_render_point_depth · tb_statistical_outlier_removal · tb_radius_outlier_removal
*Provenance: ops.py — 2D 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.