Sky and Telescope - July 2017 - 69
IMAGE AS MASK A stretched copy of the target image is used as a Deconvolution Mask in PixInsight to limit detail
enhancement to the brightest areas.
When to Apply Deconvolution
While most astrophotographers have success improving their
images using standard sharpening tools like the popular
Unsharp Mask tool in Adobe Photoshop, they often report that
deconvolution either didn't work or made their images appear
worse. I've learned that, to be applied successfully, deconvolution must be performed at the right place in the workﬂow.
The technique is best applied to linear images, before the
image is non-linearly stretched to reveal both the brightest areas and faintest regions simultaneously. Non-linear
processing alters the relationships between pixel values in the
original data, and deconvolution won't work properly after
that point. (Standard sharpening tools may work well on
these stretched ﬁles, however.)
Deconvolution works best on luminance images, which
tend to have the highest signal-to-noise ratio. A luminance
image can be created by shooting through a luminance
ﬁlter, or synthesized from the combined red, green, blue, or
narrowband photos, and can even be extracted from a color
image. While most astronomical image-processing software
includes some form of deconvolution tool, my preference is
to use PixInsight (pixinsight.com). For more information on
where deconvolution ﬁts in a PixInsight workﬂow, see S&T:
Aug. 2016, p. 66.
tool, found in the pulldown menu at PROCESS > MaskGeneration > RangeSelection.
In this process window, increase the lower limit and the
smoothness settings, then click the "apply" square at the bottom left to produce the mask. You might need to experiment
with these settings and apply them a few times until you
construct a mask that blocks everything but the stars and the
brightest area you want to deconvolve.
To perform deconvolution in PixInsight, three support images
need to be generated from the image that you want to process.
The ﬁrst support image is a deconvolution mask. Because
the background in your image contains little detail, deconvolution in these areas will only enhance noise. So you should
generate a mask that protects these areas from the deconvolution process. You can make it in varying ways, including
creating a copy of the image and stretching it using HistogramTransformation. Or you can apply the RangeSelection
MEASURING BLUR The DynamicPSF tool estimates the point
spread function (PSF) from a selection of stars in the image,
which the deconvolution tool then uses to accurately restore detail.
Choose at least 20 unsaturated stars to get the best results.
Clicking on the camera icon generates the PSF image.
s k y a n d t e l e s c o p e .c o m
* J U LY 2 017