Calibrate, Register, and Stack

Capturing hundreds of frames is only the beginning. Before stretching and color work, those measurements need to become a geometrically aligned and statistically reliable master image.

The basic pipeline is:

calibrate → inspect and reject → register → normalize → stack

Each step has a different job. Treating all of them as “stacking” makes troubleshooting harder and encourages the mistaken idea that more frames automatically repair every defect.

Lights and calibration frames

Light frames are the actual exposures of the sky. Traditional calibration uses reference frames to model repeatable signals that are not part of the target:¹

• Darks use the same exposure, gain, and sensor temperature as the lights, with no light reaching the sensor. They measure thermal signal, amplifier glow, and some repeatable pixel behavior.
• Flats record an evenly illuminated field through the same optical path and focus configuration. They describe vignetting, dust shadows, and pixel-to-pixel sensitivity differences.
• Bias frames are the shortest practical covered exposures. They model the readout offset used in some calibration workflows.
• Dark flats match the exposure and settings of the flats. They can replace bias in workflows where the flat exposure itself needs a matching dark reference.

Calibration is subtraction and division applied with care; it is not a generic cleanup filter. A poorly matched dark or changed optical path can add structure instead of removing it. Use enough calibration frames to build low-noise masters, and follow the workflow appropriate to the sensor and software.

Open the original calibration explanation

What changes with smart telescopes

Integrated telescopes hide some of the traditional setup, but they do not make calibration concepts irrelevant. The exact internal behavior can vary by model and firmware, so distinguish documented controls from assumptions.

Current Seestar documentation describes a Flat Shoot feature and a Hot Pixels Correction process.²³ The latter regenerates a thermal-noise or hot-pixel table; the documentation does not describe it as exporting ordinary bias frames. Do not infer that every saved FITS file is fully calibrated unless the documentation for the exact device and firmware says so.

For routine use:

• keep original subframes when storage permits;
• regenerate documented hot-pixel correction when the manufacturer recommends it;
• capture flats when the device provides a supported workflow and the optical path warrants them;
• and inspect the data instead of assuming automation has removed every defect.

Inspect before you combine

Reject frames for specific reasons: trails, major focus loss, cloud-softened stars, severe background changes, or failed framing. Do not discard every frame that looks less attractive when stretched individually; some defects disappear during registration and rejection, while aggressive culling can throw away useful integration.

Useful measurements include:

• star count and star shape;
• FWHM or another estimate of star width;
• background level and gradient;
• noise estimate;
• and the amount of valid overlap after alignment.

Visual inspection remains important. A numerical score cannot always distinguish faint nebulosity from a poor background or catch every satellite trail.

Registration: put the sky in the same place

Registration identifies stars and transforms every accepted sub so the same sky coordinates land on the same output pixels. It corrects intentional dithering and small framing changes, and it can rotate frames from different orientations.

Registration does not make the coverage identical. Dithering, field rotation, meridian flips, and multi-night framing create borders that are supported by fewer frames. Those edges often need to be cropped after stacking.

For filter sets or mosaic panels, use a shared reference geometry when the software permits. Hα, O III, S II, RGB, and star layers must align before they can be recombined cleanly.

Moving targets need more than star registration

A comet changes position against the star field during a session. If every frame is aligned only on the stars, the stars remain sharp but the comet’s coma and tail smear. Aligning only on the comet reverses the problem: the comet stays fixed while the stars trail.

A more complete comet workflow creates both geometries:

1. calibrate the individual light frames;
2. build a star-aligned stack;
3. register the same sequence on the comet and build a comet-aligned stack; and
4. combine the star and comet results carefully.

Siril’s official comet workflow documents separate global and comet registration, with the first and last comet positions defining its motion.⁴ This is more work than an automatic live stack, but it preserves the coma instead of asking one alignment model to follow two different motions. Open the original comet-stacking field note

Normalization: reconcile compatible differences

Transparency, Moon position, target altitude, and camera behavior can change the background and intensity from frame to frame. Normalization estimates compatible scale and offset differences before combination.

Normalization is not permission to mix fundamentally incompatible data. A frame captured through a different filter represents different wavelengths. A badly clouded frame contains information that a scale factor cannot restore. Group data sensibly by optical configuration and inspect large changes before normalizing them away.

Rejection: identify values that do not belong

Statistical rejection methods compare aligned pixel samples and reduce the influence of outliers such as satellite trails, cosmic rays, and some hot pixels. They work best with enough frames and reasonable parameter choices.

Rejection is not the same as frame curation. A long trail across a few frames may be removed statistically, while a persistent gradient or tracking smear across most frames may survive. Review the rejection maps when your software provides them: they show what the algorithm removed.

Stacking: combine the accepted measurements

Stacking combines the registered, normalized samples into a master. The result depends on:

• which frames were accepted;
• the reference geometry;
• frame weights;
• normalization and rejection settings;
• and the output method.

Average-like combination preserves quantitative relationships and improves signal-to-noise as more compatible data are added. Weighting can give cleaner, sharper frames more influence, but opaque weights should not replace judgment.

See the original distinction between stacking and preprocessing

A practical capture-to-master checklist

Before capture:

• confirm focus, framing, filter, gain, sub length, and mount mode;
• make sure dew control and storage are ready;
• decide whether the optical path or workflow needs new flats;
• and preserve enough overlap if this is a mosaic or multi-night project.

After capture:

1. Copy the source files into an untouched archive.
2. Group lights by target, filter, gain, exposure, sensor temperature, and optical configuration.
3. Build and apply matching calibration masters where required.
4. Inspect and reject only genuinely unusable frames.
5. Register to a suitable shared reference.
6. Normalize compatible data.
7. Stack with an appropriate rejection method.
8. Inspect the master and rejection maps.
9. Crop unsupported borders and save a clean linear master.

The master is the beginning of image development, not the final image. Continue with Why Processing Matters for gradients, stretching, color, stars, and finishing.

Notes and external sources

Footnotes

1. Siril, Image Calibration. Documents bias, dark, and flat correction and the noise contribution of calibration masters. ↩

2. ZWO Seestar, Flat Shoot. Current official flat-capture workflow and stated purpose. ↩

3. ZWO Seestar, Hot Pixels Correction. Describes regeneration of the thermal-noise table. ↩

4. Siril, Working with Comets. The official tutorial covers star-aligned and comet-aligned registration and stacking, plus combining the two results. ↩