Pixinsight Pipeline

• HaRGB (no luminance): disable l_stretch, l_nxt, l_bxt, lrgb_combine
• HaLRGB (with luminance): enable lum branch steps + lrgb_combine
• LRGB (no Ha): set files.Ha to "", disable ha_sxt, ha_stretch, ha_curves, ha_ghs, ha_inject. Pipeline auto-detects hasHa and skips Ha file opening/cloning.
• RGB only (no Ha, no L): set files.Ha and files.L to "", disable Ha + lum branch steps

Pipeline Architecture

Branches

Standard Processing Order

Pre-combination (Phase 0): 0a. gc_per_channel — Per-channel GradientCorrection on R, G, B, L individually (perChannel: true). CRITICAL for LRGB — different channels have different gradients. 0b. align — StarAlignment (align G, B to R reference)

Linear (on combined RGB composite):

1. combine_rgb — PixelMath R/G/B into RGB, copy astrometry
2. gc — AutomaticBackgroundExtraction or GradientCorrection on combined image
3. bxt_correct — BlurXTerminator correctOnly (fix aberrations before calibration)
4. plate_solve — ImageSolver astrometry (needed for SPCC)
5. spcc — SpectrophotometricColorCalibration
6. scnr — Green cast removal
7. bxt_sharpen — BlurXTerminator sharpening pass
8. nxt_pass1 — NoiseXTerminator linear denoise (moderate: 0.30)
9. sxt — StarXTerminator on linear (stars=true, NO unscreen)

Stars branch (non-linear extraction): 10. star_stretch — Load pre-SXT checkpoint, apply same HT+GHS, SXT with unscreen=true 11. star_saturate — CurvesTransformation S channel boost

Ha branch: 12. ha_sxt — StarXTerminator on linear Ha 13. ha_stretch — HT auto-stretch 14. ha_curves — Custom transfer curve 15. ha_ghs — GHS midtone/highlight control

Luminance branch (if L data): 16. l_stretch — HT auto-stretch 17. l_nxt — NoiseXTerminator denoise 18. l_bxt — BlurXTerminator sharpening (optional)

Main stretch: 19. stretch — HT auto-stretch + GHS refinement passes

Non-linear processing: 20. nxt_pass2 — Post-stretch denoise (stronger: 0.50) 21. curves_main — Contrast S-curve + saturation 22. ha_inject — Three-part Ha injection with nebula mask 23. lrgb_combine — LRGBCombination (if L data; LinearFit L to RGB luminance first) 24. lhe — Local Histogram Equalization (tonal separation, with lum mask + maskGamma) 25. lhe_fine — LHE with smaller radius for micro-contrast (optional) 26. hdrmt — HDRMultiscaleTransform (core detail in bright regions, with lum mask + maskGamma) 27. nxt_final — Light denoise after LHE/HDRMT (0.30 — cleans amplified noise without over-smoothing) 28. hue_boost — Hue-selective saturation (blue spiral arms, pink HII regions; galaxies only) 29. shadow_darken — Gentle background darkening via lightness curve (galaxies only) 30. channel_boost — Per-channel color correction (filter-specific residual cast) 31. curves_final — Gentle lightness/saturation refinement 32. star_add — Screen blend stars back

Key Techniques

Non-linear star extraction — Avoids halo bloating. Load pre-SXT checkpoint, apply identical stretch (HT+GHS), then SXT with unscreen=true. Screen blend to recombine: ~(~$T*~(strength*stars)).

Ha injection (three-part with nebula mask):

1. Conditional R-channel: R + strength * max(0, Ha - threshold*R) — adds Ha where it exceeds existing R
2. Luminance boost: LRGBCombination with Ha as luminance — transfers structural detail to all channels
3. Detail layer: $T + detailStr * (Ha - GaussianBlur(Ha, sigma=15)) — color-neutral filament enhancement

GHS via PixelMath — The GHS process module (.dylib) is not installed. Use the PixelMath fallback with exp(exponent*ln(base)) (no pow() in PixelMath). See GHS reference.

Checkpoint system — XISF checkpoints before heavy steps. --restart-from <stepId> to resume.

Mask gamma compression — Luminance masks for LHE/HDRMT on galaxies must use gamma compression to prevent bright cores from saturating to 1.0 in the mask. Formula: exp(gamma * ln(max(rescaled, 0.00001))). LHE uses maskGamma=2.0, HDRMT uses maskGamma=1.5.

Hue-selective saturation (hue_boost) — Replaces blanket saturation for galaxy targets. Uses PixelMath to classify pixels by hue (R/G/B ratios) and apply per-hue boost factors: blue spiral arms (1.30), pink HII regions (1.25), golden bulge (1.0). Formula: lum + factor * (channel - lum).

Per-channel gradient correction — Phase 0c applies GradientCorrection to R, G, B, L individually before combination. Essential for LRGB where different filters/nights produce different gradient profiles. Has baseline guard that reverts if GC makes any channel worse.

HDR headroom — Modified Seti Hermite HDR compress that caps maximum pixel value below 1.0. Gives HDRMT working room in bright galaxy cores. L headroom=0.10, RGB headroom=0.05.

Config Format

{
  "version": 2,
  "name": "TargetName Workflow (Iteration N - description)",
  "files": {
    "sourceFolder": "/path/to/data",
    "L": "", "R": "path.xisf", "G": "path.xisf", "B": "path.xisf", "Ha": "path.xisf",
    "haAlignCrop": { "left": 0, "top": 0, "right": 0, "bottom": 0 },
    "outputDir": "/path/to/output", "targetName": "TargetName"
  },
  "branches": { ... },
  "steps": [
    { "id": "step_id", "name": "Display Name", "branch": "main",
      "enabled": true, "params": { ... } }
  ]
}

Iteration Workflow

1. Run pipeline with current config
2. IMMEDIATELY write iteration_XX.md — this is the FIRST thing after the run completes, before any other analysis or conversation. DO NOT SKIP THIS STEP.
3. Review JPEG previews (exported at each step)
4. Identify what needs adjustment (too much/little contrast, color, noise, etc.)
5. Modify params in config JSON
6. Re-run (or --restart-from a checkpoint for speed)

Mandatory Deliverables (every iteration, no exceptions)

• iteration_XX.xisf — Full-resolution XISF (saved by pipeline)
• iteration_XX.jpg — Full-resolution JPEG preview (saved by pipeline)
• iteration_XX.md — MANDATORY markdown write-up. Must include: config used, parameter change table vs previous iteration, results (final median), assessment, issues found, and credits/software. See existing iteration_*.md files for format. Write this IMMEDIATELY after the run — do not defer.

Parameter Tuning Guide

Core Parameters

Gradient Correction

LHE / HDRMT Masks

HDR Headroom

Hue-Selective Saturation (hue_boost)

Channel Color Correction

Common Issues

• Stars have halos: Avoid star erosion/threshold — use clean non-linear extraction instead
• Over-processed look: Reduce LHE amount, disable LHE fine and HDRMT, use gentler curves
• Magenta/purple background: SPCC issue — check sensor QE and filter profiles
• PixInsight crash: Check memory (warn at 8GB). Close intermediate images aggressively.
• ImageSolver not defined: Known — #include doesn't work in eval. SPCC still works without plate solve if image has WCS.

Star Method: Linear Seti Stretch (Recommended)

Inspired by Seti Astro (Bill Blanshan). This is the recommended approach — produces tight, point-like stars without bloating.

How it works:

1. SXT on linear data (stars=true, no unscreen) — extract stars before stretch
2. Clip background pedestal from linear star residuals
3. Stretch with N iterations of the Midtone Transfer Function (MTF):

   MTF(m, x) = (1-m)*x / ((1-2m)*x + m)
   

4. Apply strong saturation boost (CurvesTransformation S channel) — the MTF stretch desaturates stars, so aggressive saturation compensation is needed
5. Add stars back with simple PixelMath addition at 100% strength (screenBlend: false)

Config params:

"star_stretch": { "starMethod": "linear", "setiMidtone": 0.20, "setiIterations": 5 }
"star_saturate": { "starSaturationCurve": [[0,0],[0.10,0.55],[0.30,0.80],[0.55,0.95],[1,1]] }
"star_add": { "starStrength": 1.00, "screenBlend": false }

Key lessons:

• setiMidtone 0.20 with 5 iterations is a good starting point
• Star saturation must be applied AFTER the Seti stretch (stretch desaturates)
• Use a very aggressive S-curve — linear stars have low inherent saturation
• Simple addition at 1.00 strength works well (not screen blend)
• Do NOT use star erosion/threshold — creates artifacts. Clean extraction is sufficient.

Alternative "nonlinear" method (legacy, not recommended): loads pre-SXT checkpoint, applies identical HT+GHS stretch, then SXT with unscreen=true. Produces slightly bloated stars.

Credits / Inspired By

Iteration Workflow — Required Artifacts

Every pipeline run MUST produce a complete set of artifacts in the target's output/processed/ folder. This is the standard way of working — follow it for every target.

Artifacts per iteration

Iteration notes template

Write iteration_XX.md IMMEDIATELY after every pipeline run (before any other work). Include:

1. Config: which JSON file, one-line description
2. Key Changes: table of what changed from previous iteration, with rationale
3. Results: per-channel GC stats, stretch metrics (median, max), shadow darken pre/post, final median
4. Assessment: wins (what improved), problems (what's still wrong), diagnosis
5. Next Steps: specific parameter changes for the next iteration

Pipeline diagram

Generate a Mermaid.js flowchart showing:

• All enabled steps as nodes with key parameters
• Branch architecture (main, stars, lum, ha) with fork/merge points
• Different colors per branch (main=blue, stars=gold, lum=purple, ha=red)
• Merge points highlighted (green)
• Disabled steps omitted

Update the diagram when the pipeline structure changes (new steps added/removed, branch changes).

Config JSON versioning

• Name configs as TargetName_vXX.json (e.g., M81_M82_LRGB_v39.json)
• Keep only the latest "best" config and the immediately previous version in the project
• Archive iteration notes for reference, but configs older than N-2 can be cleaned up
• The config IS the reproducible recipe — it must fully specify every parameter

Reference Files

• PJSR Process Parameters — LHE, HDRMT, MorphologicalTransformation, LRGBCombination, Convolution, SCNR
• PJSR Gotchas — File I/O, eval quirks, crop masks, ECMAScript 5 constraints
• Xterminator Tools — SXT, NXT, BXT parameter reference
• GHS Stretch — GHS formula, PixelMath implementation, multi-pass strategy
• Processing Knowledge — Equipment settings, quality assessment, lessons learned