stopping op• Data kinds: image2d → table
• Call: import fullseye as fs; fs.ledger.iv_richardson_lucy_until(image=None, psf_sigma=1.5, psf=None, noise='poisson', sigma=None, gain=1.0, tau=1.0, max_iter=2000, plateau_window=25, plateau_rtol=0.0001, keep_history=True, callback=None) -> 'dict' (implementation को सीधे call करना हो तो import deconvstop; deconvstop.iv_richardson_lucy_until(image=None, psf_sigma=1.5, psf=None, noise='poisson', sigma=None, gain=1.0, tau=1.0, max_iter=2000, plateau_window=25, plateau_rtol=0.0001, keep_history=True, callback=None) -> 'dict', registry से लेना हो तो opsconverge.get("iv_richardson_lucy_until"))
Discrepancy principle से रुकने वाला Richardson-Lucy deconvolution (Poisson: Bertero 2010, τ = 1; Gaussian: Morozov); model की range से बाहर के data को मना करता है।
> नीचे का विस्तृत विवरण मूल text में है — सारांश और headings का अनुवाद हो चुका है।
image: the blurred, noisy 2-D image. psf: a 2-D PSF (normalised here), or a Gaussian of psf_sigma px. The blur is periodic (FFT), so the RL step x <- x * H^T(y / H x) keeps sum(H x) = sum(y) exactly (Lemma 3.2 of Benvenuto & Piana).
• noise="poisson": image * gain must be photon counts (non-negative integers within 1e-6); stop at the first k with the Poisson discrepancy D_k <= tau (Benvenuto & Piana 2012 eq. (30); tau = 1 as proposed by Bertero et al. 2010).
• noise="gaussian": sigma (the noise standard deviation, in image units) is required; stop at the first k with ||H x_k - y||^2 / (N sigma^2) <= tau (Morozov, eq. (18)). The image must be non-negative (add the known offset first).
The iteration starts from the flat image of the data's mean (k = 0). Returns image (x_k in the input units), stopped, status, k_stop, discrepancy_history (D_0, D_1, ...), mass_rel_error_max (Lemma 3.2), tau and noise.
Not stopped -> image = None and status:
• "outside_model_range" — the discrepancy levelled off above tau (relative decrease below plateau_rtol per iteration over plateau_window iterations): y is not in the range of the model, so the criterion can never be met (Benvenuto & Piana: the rule is a regularisation iff y is in H(C)). Typical causes: the assumed PSF is wider than the real blur, a background is missing, the noise level is wrong.
• "max_iter" — still decreasing when max_iter ran out.
The last iterate is under last_iterate (not an estimate). callback(k, x_k, D_k), if given, is called after every iteration (x_k in the input units) — for plotting the run.
• convergent_computation family guide
• Sample data catalog (DL URL / license) — 2-D के लिए skimage.data (BSD/public) + synthetic, 3-D के लिए real data sources (Stanford/PDS आदि) के DL URL।
• Operators का स्रोत और संदर्भ — इस op family के मूल research/methods के स्रोत।
• Algorithm का मानक स्रोत (लेखक, वर्ष) और उपयोग ऊपर की family usage guide में दिए गए हैं।
• converge_tour — py -3.11 examples/converge_tour.py
table को input के रूप में लेते हैं)—
stopping)—
*Provenance: deconvstop.py — CONVERGE operator registry. यह per-op note tools/opdocs.py md से अपने-आप generate होता है (हाथ से edit न करें)।*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.