Description
Dark halos are a characteristic artifact of poorly dosed deconvolution.
Deconvolution is a mathematical operation that attempts to reverse the blur introduced by atmospheric turbulence and the optics: it tightens stars and brings out fine details. But like any inverse operation, it is unstable. When pushed too far, or applied without a suitable mask, it digs into the transition zones between star and sky background, creating a dark ring around each bright star.
This defect has become much more common since the widespread adoption of BlurXTerminator (RC-Astro), whose "Sharpen Stars" slider is very powerful and tempting to push too high.
In legacy PixInsight (Deconvolution process), the phenomenon is also known as dark ringing.
Visual signature
Each bright star is surrounded by a dark ring, sometimes blacker than the ambient sky background, cutting sharply against the surrounding nebulosity.
The effect is most visible on medium-to-bright stars rather than faint ones. Against a nebular background, the halos create an impression of "holes" around the stars, as if they had been cut out of the nebula with a knife.
The stars themselves often appear unnaturally shrunken, almost point-like, losing the soft gradient characteristic of a correct PSF.
Differential diagnosis
Do not confuse with dark ringing from a star-reduction tool such as StarXTerminator when the star/background separation is performed poorly (the halos appear after recombination, not after deconvolution).
Also distinct from the optical reflection halo (internal reflection inside the filter or lens), which produces a bright, not dark, halo, generally asymmetric and always in the same direction.
Black sky-background clipping can locally mimic the effect, but it affects the whole image uniformly rather than specifically the edges of stars.
Finally, a poorly configured HDRMultiscaleTransform applied to stars can produce similar dark rings.
Probable causes
- BlurXTerminator "Sharpen Stars" slider pushed beyond 0.50 without necessity
- BlurXTerminator applied twice in succession to the same image
- PixInsight Deconvolution run with too many iterations (>50) and insufficient regularization
- No star-protection mask, or a mask that is too permissive
- Poorly estimated PSF (FWHM under-evaluated) leading to excessive correction
- Regularization (Wavelet/Local Support) disabled or too weak
- Deconvolution applied on a non-linear image (must be done in linear state)
- Source image already very sharp (good seeing) where deconvolution adds nothing
Course of action
- Reduce BlurXTerminator "Sharpen Stars" to between 0.25 and 0.50 depending on seeing
- Always apply deconvolution only once, early in the linear workflow
- In PixInsight Deconvolution, enable Local Support with a well-calibrated star mask
- Measure the real PSF with DynamicPSF before Deconvolution; do not enter values by eye
- Systematically compare before/after with Blink to confirm the absence of dark rings
- If halos are already present, redo the step with reduced parameters rather than attempting to repair
- For saturated bright stars, deconvolution is pointless: mask them out entirely
- On star-dense targets (clusters, Milky Way), be even more conservative
The Doc's advice
BlurXTerminator is a bit like seasoning: always use less than instinct suggests. If you see even the faintest dark ring around your bright stars, you have already passed the optimal point. Go back, lower the slider by 0.10, and trust the more natural result. Nobody has ever criticized an image for having stars that are "not tight enough," but dark halos stand out to everyone.
Think you can see this defect in your image?
Run a diagnosisFrequently asked questions
Can dark halos already present in the image be corrected?
With difficulty. Once the information around the stars has been dug out by deconvolution, it is lost. A few workarounds exist: reducing star size with StarXTerminator then replacing them with synthetic stars, or local cloning via PixelMath on the most visible areas. But the real solution is to restart processing from the linear image with softer parameters.
Why does BlurXTerminator produce these halos more easily than classical deconvolution?
Because it is far more powerful and therefore more tempting to push. The BXT neural network automatically estimates the PSF and applies a strong correction in a single pass. Without the manual friction of adjusting PixInsight Deconvolution (iterations, regularization, masks), one quickly goes too far. The rule of thumb: if the image impresses you enormously after BXT, there is probably already a bit too much.
Should an image taken under poor seeing be deconvolved?
Yes, and that is actually where deconvolution is most beneficial. An image with FWHM above 4" benefits more from correction than a sharp image at 1.8". However, the worse the seeing, the wider the PSF and the higher the risk of artifacts: compensate with conservative parameters and a good mask, not with a more aggressive correction.
On which images is the risk of dark halos highest?
On images with high star-to-background contrast: star-rich fields with bright stars against faint nebulosity, globular clusters, Milky Way regions. Conversely, an isolated galaxy on a black background presents little risk since foreground stars are few. Adapt your parameters to the target type: the same BXT setting can be perfect on M31 and catastrophic on M11.