Combine SHO and HOO Images with NarrowbandNormalization

Narrowband filters isolate small wavelength ranges around emission lines. In this lesson, H means hydrogen-alpha (Hα), S means [S II], and O means [O III]. Each master is grayscale; assigning those masters to red, green, and blue produces a color representation of their structures.

SHO and HOO name the channel mapping in R, G, B order. SHO uses three emission-line datasets; simple HOO uses two, with O III assigned to both green and blue. Duplicating O III does not create an additional independent measurement.

These are mapped palettes, not ordinary broadband RGB color. Changing their balance changes the presentation of the emission. Record the mapping and adjustments rather than describing an aesthetic normalization as a measurement of physical line ratios.

Start with calibrated, registered, integrated monochrome masters covering the same field. Use Hα and O III for HOO, and add actual S II data for SHO. Keep untouched copies with clear identifiers such as Target_Ha, Target_OIII, and Target_SII.

Check common dimensions, orientation, crop, and alignment at 1:1. Differences in seeing, focus, or channel quality can appear as colored rims after combination. A weak S II or O III channel may need more exposure; a boost can reveal its noise as well as its signal.

This exercise starts with linear masters and uses the simple mappings below. An OSC dual-band capture requires an appropriate extraction workflow before you can treat its outputs as Hα and O III estimates. A normal RGB or HOO image does not contain a separately measured S II channel just because a palette tool can make it look like SHO.

Choose one palette for the first exercise. Keep a separate saved output if you later compare the other mapping.

Complete intended linear-only operations before creative channel balancing and permanent stretching. Use the linked sharpening/noise-reduction and star-separation lessons where appropriate. Keep the masters and the combined linear checkpoint so you can return to them.

For this exercise, make a restrained permanent stretch of the composite using MAS or another controlled method, then inspect it with STF off. Save Target_SHO_stretched or Target_HOO_stretched. Display-only STF does not turn a linear image into the stretched input used here.

A starless branch can make palette changes easier to judge without changing stars. If you choose one, preserve an aligned stars layer for later restoration. Stars may also remain in the first practice image, but inspect how normalization changes their colors and halos.

Do not copy broadband SPCC settings onto a mapped SHO/HOO image as a shortcut to a preferred palette. Physical narrowband calibration and creative normalization have different purposes. Likewise, MGC reference bands must match the actual channel content and available coverage, not merely the labels R, G, and B.

NarrowbandNormalization is published by Cosmic Photons in collaboration with Bill Blanshan. It balances intensities in a narrowband color image. Install it using the linked setup guide and use its built-in help for the current parameter definitions.

The palette selection must describe the data already in the input image. Selecting SHO does not acquire sulfur data or automatically repair an HOO input. Use a fresh copy of the stretched checkpoint for each trial.

In SHO, first decide which structures are too dominant and which are hard to see. Try a small change to one available channel boost while keeping the other controls fixed. If the weaker channel is mostly noise, a stronger color is not a better measurement.

In HOO, assess the relationship between hydrogen-dominated regions and oxygen structure. Do not add an S II boost to simulate missing observations. The palette may offer different active controls because the input contains only two independent line datasets.

Keep SCNR off while learning what the channel balance alone does. If you later use green suppression for a chosen presentation, compare a small amount separately and inspect whether it removes useful color variation. Green is not automatically an error in a palette that assigned an emission line to green.

Lightness choices and shadow, highlight, and brightness adjustments can change more than the overall hue. Explore them as separate experiments, watching faint emission and bright cores. Do not push several sliders simply to imitate the gold/blue appearance of another target.

When the preview is useful, apply once to the working duplicate and save a trial such as Target_SHO_NBN_trial01.xisf. Inspect the actual output with STF off. Confirm the preview and applied result were evaluated on the same image and state.

Compare with the pre-normalization checkpoint at matched zoom. Inspect bright boundaries, subtle filaments, dark lanes, faint background structure, stars, and halos. Look for hard color transitions, clipped cores, amplified color noise, or a nearly uniform palette that hides differences present in the channels.

Return to the same stretched checkpoint for each settings comparison rather than repeatedly normalizing the previous trial. A second pass on an already altered palette is not the same experiment as changing the first pass’s settings.

There is no requirement for equal amounts of each display color. Retain the balance that preserves useful structure and suits the stated purpose of the image, or retain the original if normalization does not help.

After accepting a broad palette balance, use the ColourMask and CurvesTransformation lesson for restrained local adjustments. Build masks from the current accepted image and inspect their coverage, since earlier masks may no longer select the same colors.

If working starless, restore a prepared, aligned stars branch using the ImageBlend lesson. Broadband RGB stars can provide a separately calibrated broadband color presentation when available. Narrowband stars remain limited by the emission-line data; normalization does not recover their missing broadband spectrum.

For the first exercise, keep the selected starless result, stars source, and combined image separately. Avoid applying another global palette change to the star-restored image unintentionally.

SHO is overwhelmingly green: verify the channel assignments, then assess channel signal and the initial stretch. Compare a controlled balance adjustment before suppressing all green.

The palette looks wrong after normalization: confirm that the selected palette describes the input mapping and that the input is at the intended processing state. Start again from the labeled checkpoint rather than guessing which channels were used.

A boost creates blotchy red or blue regions: inspect the relevant original master for low signal, residual gradients, or noise. Smaller boosts and better data may be more useful than stronger smoothing.

Colored star rims or doubled structure: check registration and profile differences. Palette balancing cannot align mismatched channels.

A result has little color distinction: check whether the channel data genuinely differ, whether some input was duplicated accidentally, or whether clipping and strong adjustments removed variation. In simple HOO, G and B intentionally share the O III source.

LinearFit, LocalNormalization, and NarrowbandNormalization are not interchangeable. Matching levels for another processing task is not the same as choosing the displayed narrowband palette, and none creates missing emission-line measurements.

Save the retained result, the pre-normalization image, and the process instance. Record the exact R/G/B mapping, input state, stretch, lightness choice, channel boosts, SCNR amount, and other adjustments.

When sharing the image, identify it as SHO, HOO, or the actual modified mapping. Be clear when additional mixing or synthetic channels were used. The mapping and processing choices help viewers understand what its colors represent.

The publisher page links a demonstration by Bill Blanshan; use it and the installed process documentation for version-specific controls. This original beginner sequence uses a stretched-composite workflow and example baseline choices, not a universal normalization prescription for every dataset.