Cosmetic Correction

Identify and replace isolated defective pixels or very small residual defects without treating stars, compact structure, or ordinary noise as defects.

schema 1 · revision 1

Recognize it

Stable hot or cold pixels recur at sensor coordinates in calibrated subframes.
Sparse one-pixel colored specks survive integration and become conspicuous after stretch.
Impulse defects would bias later denoise or sharpening operations.

Intended output

Calibrated subframes corrected from a master-dark or bad-pixel map when possible; otherwise a documented sparse residual cleanup whose corrected-pixel count and star preservation have been checked.

Required input state

Prefer calibrated, pre-registration subframes plus a matching master-dark or bad-pixel map.
For residual post-stack cleanup, use a linear integrated image and a sparse-defect hypothesis.
Record CFA/debayer state because CFA-aware detection differs from RGB-image correction.

What NOVA does

NOVA's post-stack residual cleanup operates per channel with a 5x5 local median, flags absolute residuals above 5 times the median absolute residual, keeps only connected components of at most three pixels, and replaces them with the local median.

Use when

A master-dark or bad-pixel map identifies repeatable sensor defects before integration.
A linear stack contains sparse, isolated residual hot/cold pixels that survived rejection.

Skip when

The candidate defects form stars, compact target structure, trails, or extended regions.
The correction count or fraction is unexpectedly large.
Blink/difference inspection shows real signal changing.

Limits

Ordinary cosmetic correction does not repair gradients, broad artifacts, trails, or bad calibration.
Post-integration cleanup is not equivalent to correcting every calibrated subframe before registration and stacking.
Aggressive thresholds can replace star cores and real compact structure.

Scientific and aesthetic notes

Pre-integration correction preserves the stacking engine's opportunity to reject and weight data correctly.
A successful tool execution proves procedure, not that the chosen stage or threshold was appropriate.
Judge sparse defects at native scale and inspect a difference image; a full-frame beauty slider can hide one-pixel damage.

Evidence and measurements

after cosmetic correction before and after

beforeafter

cosmetic correction before and after

Cross-tool matrix

Tool	Equivalence	Expected result	Evidence
nova	Exact replay	Only isolated components of at most three flagged pixels are replaced by local medians.	nova-cosmetic-source
pixinsight	Functional alternative	Repeatable isolated defects disappear from subframes without changing stellar profiles.	m66-manual-verification
siril	Functional alternative	Sparse hot/cold pixels are replaced without an obvious broad change or measurable compact-signal damage.	siril-calibration-1.4.4, siril-cosmetic-1.4.4, m66-manual-verification
saspro	Functional alternative	Sparse isolated hot and cold pixels are replaced during light-frame calibration without changing real compact signal.	saspro-source-1.18.0, saspro-synthetic-2026-08-14

NOVA Python pipeline 1.24.7

Run on the linear stack only when sparse residual defects remain after integration.
Use the standing code and ontology defaults for sigma, odd median-kernel size, and maximum connected-defect size.
Review pixels_fixed, pct_fixed, a native-scale blink, and a difference image before accepting.

Controls and starting ranges

sigma: 5.0 standing ontology and implementation default
kernel_size: 5 pixels, odd; standing ontology and implementation default
max_defect_size: 3 connected pixels; standing implementation default

Mask behavior: No user mask; the connectivity filter is the implementation's structural safeguard.

Failure modes

Threshold is low enough to flag star cores.
Connected real structure fragments into small flagged islands.
A large correction fraction is accepted without investigation.

Recovery

Revert to the unchanged linear input and raise the threshold or correct subframes before integration.

PixInsight CosmeticCorrection exact version not recorded; procedure tool-tested

Prefer applying CosmeticCorrection to calibrated subframes before registration and integration.
Use a master-dark defect list when available; otherwise configure conservative hot/cold detection.
Inspect the corrected subframes or rejection evidence at native scale before continuing.

Controls and starting ranges

hot/cold sigma: 3-5 is a tested range; tune from defect and star-preservation evidence
input: calibrated subframes; post-stack use is residual cleanup only

Mask behavior: Tool-specific defect lists/auto-detection; no handbook claim of ordinary image-mask behavior.

Failure modes

Using the batch process successfully on an integrated stack and calling the recipe stage validated.
Thresholds remove faint-star cores or compact detail.

Recovery

Undo and move the correction to calibrated subframes or use a more conservative defect map.

Siril calibration or Cosmetic Correction 1.4.4 documentation; standalone procedure tool-tested in 1.4.2

During calibration, prefer master-dark sigma detection or a bad-pixel map; enable CFA handling for Bayer data.
Estimate the corrected count before batch processing; Siril flags estimates above one percent for review.
For an already debayered residual stack, use the standalone Cosmetic Correction filter with CFA disabled and inspect the reported count.

Controls and starting ranges

cold sigma: 5.00 tested starting point; tune from the corrected-pixel evidence
hot sigma: 5.00 tested starting point; tune from the corrected-pixel evidence
amount: 1.00 tested starting point
CFA: on only for CFA/Bayer input; off for debayered RGB
review threshold: investigate when the calibration estimate exceeds 1%

Mask behavior: Siril documents cosmetic correction as full-image sensor-defect processing, not a mask-limited operation.

Failure modes

CFA state is set incorrectly.
A high corrected count is accepted without revisiting thresholds or calibration.
Standalone post-stack cleanup is represented as equivalent to pre-integration correction.

Recovery

Undo, correct the CFA setting, raise sigma, or return to sequence calibration with a master-dark/bad-pixel map.

Seti Astro Suite Pro Stacking Suite Cosmetic Correction 1.18.0 source-inspected and synthetic-array tested

Enable Cosmetic Correction while calibrating light frames in the Stacking Suite, before registration and integration.
Use Bayer-aware correction for undebayered CFA data and ordinary mono/color correction only for the matching input state.
Start conservatively, then inspect corrected frames and a difference image for repaired defects and altered star cores.

Controls and starting ranges

recommended hot sigma: 5.0 in the installed Stacking Suite
recommended cold sigma: 5.0 in the installed Stacking Suite
input mode: Bayer pattern for CFA data; debayered/mono path otherwise

Mask behavior: The Stacking Suite selects a Bayer-aware or debayered/mono full-frame algorithm; defect handling is not an ordinary image mask.

Failure modes

The wrong Bayer pattern or debayer state mixes unlike color samples.
Thresholds classify a sharp stellar core as a hot pixel; the synthetic check showed this is possible and requires inspection.
Interactive Blemish Blaster is mistaken for repeatable calibration-stage cosmetic correction.

Recovery

Disable or raise the relevant threshold and recalibrate from the unchanged light frames.

Validation status

Not yet Jeff-validated. Use this as sourced guidance, not a certification.

Sources

M66 manual recipe verification feedback M66_Manual_Recipe_Verification_Feedback.md — Jeff validated
NOVA isolated-defect correction implementation and ontology defaults nas_server/seti_astro.py:3133; nas_server/processing_ontology.json:325 — NOVA source-confirmed
M66 workflow 1.24.7 execution record critiques/20260802_045251_seestar_galaxy.json — NOVA execution record
Siril 1.4.4 calibration documentation — Vendor-documented
Siril 1.4.4 cosmetic correction documentation — Vendor-documented
Installed Seti Astro Suite Pro 1.18.0 processing source saspro/pedestal.py; saspro/numba_utils.py; saspro/stacking_suite.py; saspro/abe.py; saspro/sssc.py — Artifact-confirmed
SASpro 1.18.0 synthetic-array operation checks 2026-08-14 local validation: pedestal, cosmetic correction, and ADBE — Tool-tested

Related evidence

The M66 recipe and Jeff-validation ledger will link here when their static exports are available. Until then, this article remains explicitly unvalidated.