21  Mono, LRGB, and Narrowband

A mono camera records no color of its own: you shoot the same target through different filters and combine the results afterwards. This chapter is the mono-and-filter-wheel branch of the Polaris pipeline: how to take the per-filter masters that come out of STUDIO’s Integrate step (Chapter 20) and turn them into a single color image, without leaving Polaris. If you shoot OSC (one-shot color, a camera that captures color directly), your path is the end-to-end workflow in Chapter 26; you only need this chapter for dual-band palette tricks.

One point up front: luminance is optional. Plenty of mono workflows are pure RGB, with no separate luminance pass, and narrowband-only shooters skip RGB entirely. Use the RGB mode when you have R, G, and B masters; the LRGB mode when you also captured a luminance master (shot through a clear or near-clear filter) and want its higher signal-to-noise ratio to drive the image’s lightness; and the narrowband palette or PixelMath modes for everything else.

21.1 What you need before starting

• A mono camera and filter wheel.
• At least two per-filter masters already produced by STUDIO’s Integrate, one master_*.fits per filter, landing under {rig}/integrated/{target}/{filter}/. RGB needs 3 masters, LRGB needs 4, and PixelMath needs as many as your formula references (minimum 2). If you have not done the per-filter integration yet, Chapter 20 covers it.
• Masters stretched to comparable brightness ranges. In practice: either keep them all linear (the raw stack output), or pre-stretch each one in STUDIO with the same parameters. Polaris trusts that the channels are comparable; if one is wildly brighter than the others, the output shifts toward it.

A typical mono session lands one of these layouts on disk:

# Plain RGB (no luminance), common for fast or short sessions
{rig}/integrated/M81/
  R/master_R_30x180s.fits
  G/master_G_30x180s.fits
  B/master_B_30x180s.fits

# Classic LRGB (luminance master adds SNR)
{rig}/integrated/M81/
  L/master_L_30x180s.fits          (extra L pass for noise control)
  R/master_R_15x180s.fits
  G/master_G_15x180s.fits
  B/master_B_15x180s.fits

# Narrowband only (no RGB at all)
{rig}/integrated/NGC7000/
  Ha/master_Ha_20x300s.fits
  OIII/master_OIII_15x300s.fits
  SII/master_SII_10x300s.fits      (optional, for SHO palette)

21.2 Opening the Combine dialog

In STUDIO, add the per-filter masters to the Lights slot of the Stack workspace (or multi-select them in the frame grid with Ctrl-click; the selection bar shows the count so you can verify the pick), then click Combine channels. The dialog first asks for the combine mode, then collects the role of each file.

21.3 The combine modes

RGB

The default, and the one most mono shooters use: three mono masters in, one RGB FITS out. No luminance master required; this is the right mode when you only captured RGB filters or chose not to spend session time on a separate L pass.

Polaris auto-fills the R, G, and B roles from each frame’s FILTER header. If your filter names do not match (you set FILTER=Red, say), pick them manually from the dropdowns.

Output: {rig}/integrated/{target}/composed/rgb_{target}_{stamp}.fits

LRGB

For shooters who also captured a luminance master: four masters in (R, G, B, L), one color image out that takes its lightness from L. The luminance is where the signal-to-noise lives, because it is captured through the broadest filter; that is what makes LRGB look cleaner than plain RGB when you have it. If you have no L master, use the RGB mode instead; it gives the same three-channel compose without forcing an L slot.

Pick an algorithm:

• Lab swap (the default, recommended). Converts the RGB stack to the CIE Lab color space (which separates lightness from color), replaces the lightness channel with the L master, and converts back. Best color preservation; matches the output of PixInsight’s LRGBCombination tool (Pleiades Astrophoto, n.d.).
• Ratio (classical, faster). Multiplies R, G, and B per pixel by L / luminance(RGB). Slightly cheaper, but it can shift saturation where the ratio swings far from 1.

Output: {rig}/integrated/{target}/composed/lrgb_{target}_{stamp}.fits

Narrowband palettes

Narrowband filters isolate the light of specific emission lines (hydrogen-alpha, oxygen-III, sulfur-II), and a palette maps those masters to display colors. The built-in palettes:

• SHO (the “Hubble” palette): SII to red, Ha to green, OIII to blue.
• HSO and HOS: the same three inputs in other orders.
• HOO bicolor, for two-filter data: Ha to red, OIII to both green and blue.

Per-channel normalization matches the three backgrounds so no single filter dominates the color. Output lands in the same composed/ folder with an nb_ prefix.

Continuum subtraction

Continuum subtraction isolates the emission-line signal in a narrowband master by removing a scaled broadband master:

NB' = max(0, NB - k * Continuum)

Assign the NB (narrowband) and C (continuum, broadband) roles; the scale k is auto-estimated from the bright star pixels (the median NB-to-continuum ratio where the signal is pure continuum), or entered manually in the range 0 to 4. Stars largely cancel while the nebulosity remains. Output uses a cs_ prefix. For this mode, per-channel normalization defaults to off so the automatic k estimate is not skewed by a pre-scaled background.

PixelMath

For everything the fixed modes cannot express: custom palettes, synthetic luminance, channel arithmetic. PixelMath is a small formula language evaluated per pixel.

Each row in Variable bindings maps a selected frame to a name of your choosing; the name defaults to the filter header, so Ha and OIII masters auto-populate as Ha and OIII. Each Output expression is a formula using those names: for RGB output you write three, one per output channel, or tick Mono output for a single expression and a single-channel FITS.

Supported syntax:

• Numbers: 0.5, 1.2e3
• Operators: +, -, *, /, ** (power, right-associative)
• Parentheses: (R + G) * 0.5
• Functions: min, max, abs, pow, sqrt, exp, log, clamp(value, min, max)

Common recipes:

What you want	Variable bindings	Expressions
HOO bicolor (Ha + OIII)	Ha, OIII	R = Ha, G = 0.5*Ha + 0.5*OIII, B = OIII
SHO Hubble palette	Ha, OIII, SII	R = SII, G = Ha, B = OIII
Synthetic luminance from RGB	R, G, B	(Mono out) L = (R + G + B) / 3
Ha-boosted red for LRGB	R, Ha	R = 0.7*R + 0.3*Ha (then re-combine as L+R’+G+B)

A typo in a variable name surfaces immediately in the dialog as a job error, before any pixels are read. Division by zero falls back to 0, so no invalid values propagate into the output.

Output: {rig}/integrated/{target}/composed/pm_{target}_{stamp}.fits

21.4 Cross-channel registration: leave it on

The Register channels checkbox at the bottom of the dialog is on by default, and it is what saves you from colored fringes on every star.

The reason is subtle: each per-filter master comes out of the batch stacker aligned to its own reference frame, the best-HFR frame within that filter’s stack. So the R master is aligned to its own best R sub, and the G master to its own best G sub. They are not aligned to each other, even when shot the same night, because the references are different frames captured under slightly different seeing (the constantly shifting blur of the atmosphere). Without registration you get colored halos around bright stars (R offset by a pixel or three from G and B), ghosting of fine nebula structure, and a generally soft look.

With it on, Polaris detects stars on each input (at a detection threshold of 7 sigma rather than the usual 5, because masters have a much higher signal-to-noise ratio and a lower threshold would flood-detect star halos), matches them against the reference channel (L for LRGB, otherwise the first input), computes an affine transform per channel (a shift-rotate-scale mapping), and resamples. The transforms are recorded in the output FITS headers (REGREF, REG_R, REG_G, REG_B) for reproducibility.

Turn registration off only when you know the masters are already aligned: a permanent pier observatory with no re-framing between filters and the same reference frame across stacks. For backyard or mobile rigs, leave it on.

21.5 Per-channel normalization: also on

Filters captured across different sessions can sit on very different sky backgrounds. The Normalize per-channel checkbox scales each input so their medians line up before the combine, which keeps a bright-background R master from red-casting the output just because R was shot under a brighter sky. It defaults to on for RGB, LRGB, and the narrowband palettes; turn it off for PixelMath when you want raw channel values feeding your expression, for example when you computed your own pre-scale factors.

21.6 Run

Click Combine. Progress steps through loading, registering, normalizing, composing, and writing. Time depends on input size and how many channels need registration; on a Raspberry Pi 5:

• 2 channels, registration off: about 5 s
• 4 channels (LRGB), registration on: about 25 s (registration adds roughly 15 s of star detection and matching across the non-reference channels)
• 6 channels (LRGB plus narrowband): about 40 s

A toast confirms the output path, and the new file appears in the frame browser after the automatic rescan with a 3-channel badge (or 1-channel for PixelMath mono output).

If you find yourself repeating the same combine plus post-processing recipe, the Auto Workflow presets (Mono LRGB, Mono SHO, and OSC dual-band SHO) wrap this combine step and a full color pipeline into one saved, re-runnable workflow (Chapter 20).

21.7 After the combine

Run Color calibration on the composed master next: background neutralization as a quick first pass, a manual white reference when a tint remains, or full photometric calibration against a star catalog. Chapter 22 covers all three.

Downstream, the composed RGB or LRGB FITS is just a frame, and everything treats it like any other master:

1. GraXpert AI cleanup (background extraction, denoise, deconvolution; Chapter 23). The pipelines iterate per RGB plane automatically, so they work on a composed file exactly as they do on OSC data.
2. The EDITOR (Chapter 25) for tone work; RGB composed files open as 3-channel sessions with the color sliders active.
3. Export as JPG, PNG, or 16-bit TIFF.

21.8 What the output records

Every composed FITS carries its recipe in custom headers: CHCOMBINE (the mode), REGISTER (on or off), REGREF (the reference channel), REG_R/REG_G/REG_B/REG_L (the affine transform applied to each channel, as the six numbers M00,M01,M10,M11,Tx,Ty), NORMLIZE (on or off), and NCHANNEL (input count). Any FITS header inspector, such as PixInsight’s FITS Header view, shows them, so you can audit exactly how any composed file was produced months later.¹

21.9 Worked example: M81 pure RGB

The simpler path, and what most backyard mono shooters do. Pi 5 host, ZWO ASI2600MM Pro, EQ6-R Pro mount, Askar FRA600, RGB filter set, no luminance pass.

Step	Time	Output
Capture: 20R + 20G + 20B at 180 s each, unattended	3 h	60 raw lights
STUDIO: calibrate and integrate per filter	8 min	3 per-filter masters
Combine: RGB mode, Register on, Normalize on	15 s	rgb_M81_*.fits
GraXpert background extraction	30 s	_bge.fits
GraXpert denoise	90 s	_denoise.fits
EDITOR tone work	8 min	sidecar saved
Export JPG at quality 92	3 s	final image

Active human work: about 20 minutes. The rest is camera and Polaris time.

21.10 Worked example: M81 LRGB with an Ha boost

The same hardware plus an L filter and an Ha narrowband filter.

Step	Time	Output
Capture: 30L + 15R + 15G + 15B + 20Ha, unattended	8 h	95 raw lights
STUDIO: calibrate and integrate per filter	12 min	5 per-filter masters
Combine: LRGB mode, Lab swap, Register on	25 s	lrgb_M81_*.fits
GraXpert background extraction	30 s	_bge.fits
GraXpert denoise	90 s	_denoise.fits
GraXpert deconvolution (objects)	2 min	_decon_objects.fits
EDITOR tone work	10 min	sidecar saved
Optional: re-combine with PixelMath R = 0.7*R + 0.3*Ha	25 s	pm_M81_*.fits
Then re-combine that output with the L master, LRGB mode	25 s	Ha-boosted LRGB
Export JPG quality 92, 50% resize	3 s	final image

The Ha boost is optional, and many users run straight LRGB. But that PixelMath round-trip is exactly the kind of step that used to mean exporting to PixInsight, doing the math there, and loading the result back; here it stays in one application.

21.11 Common pitfalls

• Colored star halos in the output: registration was off. Re-run with Register channels checked.
• Output much brighter or dimmer than expected: the L and RGB brightness ranges drifted apart. Re-stretch each master in STUDIO with the same parameters before combining, or use the Ratio algorithm, which is less sensitive to absolute brightness.
• “Could not register channel X”: that filter’s master has too few detectable stars (very short total exposure, a very narrow filter, bad framing). Increase that channel’s total exposure, or turn registration off if you trust the existing alignment.
• PixelMath says “unknown variable Ha2”: a typo. The variable names come from the Variable bindings rows, and expressions must use those exact names, case-sensitively.
• The Combine button stays disabled: not enough channels assigned. RGB needs all 3 roles filled, LRGB all 4, and PixelMath needs at least 2 binding rows and a non-empty expression for every output channel.

Pleiades Astrophoto. n.d. PixInsight: Advanced Image Processing Platform for Astrophotography. Https://pixinsight.com/.

---

1. For integrators: the dialog posts to /api/studio/combine, and the server’s ChannelCombineService dispatches to an RGB compose, LrgbCombiner (Lab swap or Ratio), the narrowband and continuum implementations, or PixelMathEvaluator, which compiles each output expression once and evaluates it per pixel. The narrowband palette and continuum math are implemented from scratch, inspired by the narrowband tools in SASpro and PixInsight.