Common Myths About Integration Rejection Algorithms

Which rejection algorithm should you use when integrating? There’s a rule of thumb going around online — “pick your rejection method by frame count” — and many tutorials copy it verbatim. But this rule actually harbors quite a few myths, and this article shares my take.

The Common Rule of Thumb

The “rules of thumb” circulating online go roughly like this:

• If you have fewer than 15–20 frames, use median integration and don’t do sigma clipping.
• If you have more than 15–20 but fewer than 40–50 frames, use average integration with outlier rejection, kappa set to 2.5, and 2 iterations.
• If you have more than 40–50 frames, use average integration with outlier rejection, kappa set to 3 (or higher), and only 1 iteration.

My View

This “pick your rejection method by frame count” knowledge is actually a bit outdated, and the problems are:

1. Confirming image quality first is the real answer. If quality isn’t controlled, whatever rejection method you pick is pointless.
2. Every rejection algorithm has its own strengths and weaknesses, so judging by frame count alone isn’t accurate.
3. If you pick the rejection method by frame count, what do you do when there are very few frames? The framework itself can’t cover every scenario.

A Few Additional Notes

• I “don’t recommend” taking the rules above wholesale (the original source is a relevant passage on the APP forum); quite a few of the parameter settings are actually problematic.
• Reject kappa splits into high and low; barring special needs, I don’t set the two to the same value.
• I’ve never used median integration myself, because I’ve already checked my images for quality beforehand.

In summary: rather than agonizing over “how many frames go with which rejection method,” first get your image-quality control right, understand the characteristics and applicable scenarios behind each rejection method, and then adjust according to your actual data.