When and How to Use DrizzleIntegration

Drizzle combines information from multiple exposures sampled at different positions. It can help reconstruct undersampled structure and avoid some effects of ordinary registration interpolation. Enlarging one completed stack with Resample does not provide those additional measurements.

Drizzle cannot recover detail absent from every exposure or repair poor focus, guiding, calibration, or saturated cores. A larger image is not proof of increased resolution. Use an ordinary integration as a baseline and keep drizzle only when the comparison supports it.

This lesson uses WBPP to manage a first drizzle experiment. Manual DrizzleIntegration is also possible, but needs the corresponding registration, normalization, rejection, and source-file relationships to remain consistent.

Start with a useful set of well-calibrated, sharp exposures of the same field, with substantial overlap and varied subpixel sampling. Dithering intentionally changes the pointing between exposures. Many frames at essentially identical sampling positions do not supply the same coverage as a well-dithered sequence.

Inspect small stars in individual exposures. If their profiles are represented by very few sensor pixels, higher-scale reconstruction may be worth testing. If seeing and optics already spread stars across many pixels, increasing output scale may mostly increase processing and storage costs.

There is no universal frame-count threshold that guarantees success. The number of useful exposures, their positions, coverage, rejection losses, and chosen drop size all matter. Repeated exposures with poor offset diversity can still leave a problematic reconstruction.

Keep calibration and frame selection sound first. Do not retain trailed or cloud-damaged frames simply to make the drizzle input count larger. Confirm the intended dither pattern from acquisition records and the actual registration shifts where available.

Separate the question of a finer output grid from the question of reconstructing color-camera data.

At scale 2, width and height are roughly doubled, so the output has about four times as many pixels. Memory, disk space, and later processing costs increase; exact resource use depends on the run. Do not choose scale 3 merely because it is available.

Drop shrink controls the footprint used when distributing input samples to the output grid. Smaller drops can reduce smoothing but demand stronger sampling coverage and can expose gaps or uneven texture. Start with the installed default for the pilot and record it rather than copying an aggressive value from another dataset.

For monochrome data, drizzle each filter’s accepted exposures into its own master on the intended common reference grid. Do not mix R/G/B or S II/Hα/O III frames into one grayscale stack.

For a Bayer color camera, CFA drizzle reconstructs color from calibrated mosaic samples using their CFA pattern and geometric mapping. It is different from simply drizzling an already debayered RGB image with CFA disabled. Use WBPP’s documented CFA-drizzle path for your version.

The pipeline may create debayered views to measure stars and determine registration while the CFA reconstruction still refers to calibrated mosaic sources. Do not remove intermediate products merely because one stage appears to duplicate another, and do not enable CFA drizzle on an RGB master or monochrome data.

Verify the camera pattern and metadata. Wrong CFA information can create color artifacts that no later saturation adjustment will fix. This lesson assumes supported Bayer data; other sensor layouts require their own documented workflow.

Use a representative set with useful dither coverage and enough exposures for the chosen rejection method. A tiny pilot can verify the setup, but may not show the quality possible with the full dataset.

Drizzle-data files contain relationships and information used to reconstruct the image; they are not self-contained copies of the exposures. Preserve the referenced source images and relevant intermediate products in their existing locations while testing.

In a manual workflow, StarAlignment generates drizzle data for the input exposures, and the appropriate integration stages add rejection and other applicable information. DrizzleIntegration consumes that prepared set. Follow the installed documentation for file types and normalization associations in your version.

Do not feed the final combined master into DrizzleIntegration expecting it to recover the exposure sequence. Do not crop or alter source frames after their geometry has been recorded. Moving files or mixing data from different runs can break paths or make their metadata inconsistent.

When changing accepted frames, reference geometry, or rejection settings, regenerate affected products rather than reusing a stale drizzle set. WBPP recovery and reuse controls need the same care; check the log instead of assuming it rebuilt everything.

Display both integrations with comparable STF settings, framing, and brightness. A scale-2 image at 1:1 is more magnified than the original-grid baseline at 1:1. Compare equivalent sky regions and presentation sizes, then inspect the native output pixels separately.

Look for better star sampling or useful structural separation, while checking smooth sky, faint signal, and bright-star edges. Inspect any coverage or weight products available in your version and check edges for weak coverage.

Reject a result with unexplained grids, holes, checkerboard color, or patchy noise. Do not mistake interpolation texture for real detail. Numerical noise and SNR estimates can respond differently to changed sampling and correlated noise; do not choose a winner solely from one score.

If the pilot is sound, run the full accepted dataset and repeat these checks. Prefer the standard integration when the larger reconstruction brings no useful improvement for the intended image.

Missing source-file errors: check the original drizzle-data paths and whether intermediates were moved or deleted. Regenerate a consistent run rather than guessing replacement files.

Holes or uneven texture: review dither diversity, accepted frame count, rejection losses, output scale, and drop shrink. Compare a less demanding scale or larger drop from the same source set before blaming later stretching.

Unexpected color patterns: verify CFA input state, pattern metadata, registration, and the actual source files used. A normal overall color cast is different from a reconstruction artifact.

No useful improvement: inspect whether the data was already well sampled or limited by focus and seeing. Extra pixels and processing time are not reasons to keep a weaker result.

Save the accepted linear master, ordinary baseline, settings, logs, and required reconstruction sources. Continue to alignment checks and a common crop. Use the correct effective image scale for later plate solving; at scale 2 the nominal arcseconds per output pixel is halved, subject to any other geometry changes.

This is an original beginner experiment. The research paper explains the reconstruction principle; older PixInsight examples explain pipeline concepts. Installed WBPP and DrizzleIntegration help governs current controls, supported paths, and file relationships.