Star Reduction

Step: Finish

Makes the stars smaller so that the nebula or galaxy behind them comes into view. The module finds the stars, covers each one with a soft mask and shrinks the stars inside the mask with a minimum filter ("erosion"). Everything outside the mask stays exactly as it was. The brightest part of a star is cut away at its rim, so a star looks smaller and fainter, and the structures that were lost in its glare show up.

Parameters

Parameter	Range	Default	Effect
Strength	0…1	0.5	How much of the shrunken version is mixed into the image inside the mask. 0 leaves the image exactly as it is, 1 uses the shrunken version in full.
Mask growth	0…10 pixels	2	Extra pixels added to the radius of every star's mask. Raise it if halos or the faint rim of the stars are left behind, lower it to protect the surroundings of the stars.
Erosion size	1…3 pixels	1	Radius of the minimum filter, about how many pixels the edge of a star moves inwards. 1 is gentle, 3 shrinks the stars strongly.

When to use it

Use it on the stretched image, when the stars have grown large and bright next to the faint structures. It also works on linear data, but there the stars are small and the stretch makes them large again. Do it beforeUnsharp Mask, so that the sharpening does not lift the star edges again.

• Start with the defaults and look at the image at 100 %. Raise Strength until the stars are as small as you want them. A strength of 1 shrinks every star as far as the Erosion size allows.
• A larger Erosion size shrinks the stars more than a larger strength, but it also works a little more on the noise in the mask: the minimum of several noisy pixels is darker than their average, so in a noisy image the mask area becomes slightly darker. Reduce the noise first with Noise Reduction.
• Only stars that are found are changed. Saturated stars with a flat top, stars merged with a neighbour, stars at the border of the image and stars next to a pixel without data are not detected and stay as they are.
• At most the 1000 brightest stars are processed; fainter stars in very dense star fields stay as they are.
• Bright, compact objects that look like stars, such as the core of a galaxy or a small knot in a nebula, can be taken for stars and are shrunk too. Check the result at 100 % and lower the strength if they suffer.
• The live preview works on a reduced copy of the image, in which the mask growth and the erosion size are scaled down too. Judge the final values on the full-size result.

How it works

1. Find the stars. The same detection as for the alignment of stacking frames: the background and the noise are measured per tile, groups of connected pixels above background plus five times the noise are candidates, and those that are too small, too large, flat-topped, elongated or cut off at the border are discarded. Each star comes with its position and its FWHM (full width at half maximum). A color image is searched in the mean of its channels, so all channels share one mask.

2. Mask. Every star becomes a disc with the radius

   radius = 1.5 · FWHM + Mask growth

   A Gaussian star has fallen to about 0.2 % of its peak at 1.5 times its FWHM, so the disc takes in the whole visible star. The union of the discs is blurred with a Gaussian that is about a third of a star's FWHM wide (at least one pixel), so the mask has no hard edge: it is 1 inside, falls to 0.5 at the rim and is exactly 0 beyond the reach of the blur.

3. Erosion. In the mask, every pixel is replaced by the smallest value in a disc of Erosion size pixels around it (the minimum filter of mathematical morphology, see J. Serra, Image Analysis and Mathematical Morphology, Academic Press 1982, and R. C. Gonzalez and R. E. Woods, Digital Image Processing). The profile of a star falls steadily from its center, so the minimum over the disc is the value one disc radius further out: the star shrinks by that many pixels on all sides and loses height. Every color channel is eroded on its own.

4. Blend.

   result = pixel + Strength · mask · (eroded − pixel)

   Where the mask is 0, nothing is computed and the pixel stays exactly as it was.

Pixels without data. Pixels without a value (for example the borders left by stacking) stay without a value and are left out of the minimum.

The file information (exposure time, filter and so on) is kept.

Errors

• The image is an undebayered mosaic; star reduction would mix its colors. The pixels of the raw sensor mosaic differ in brightness from their neighbours because of the color filter, and the minimum filter would mix these colors. Run Debayer first.