Acquisition

The images were acquired with the diminutive William Optics Star 71 5-element astrograph. It was almost comic fixing this to the substantial Paramount MX mount. The heavy QSI camera is a significant load for the focus mechanism, however, and it took some time to find the best focuser clamp force to reduce flexure and still allow the focuser’s stepper motor to work. The outcome was a compromise and the asymmetrical reflections are evidence of some remaining focuser sag. The bright star Alnitak (the left-most star of Orion’s belt) challenges most refractor optics and this little scope was no different. In this case, a complex diffraction pattern around this star is the consequence of minute irregularities in the optical aperture and blue reflections from the sensor glass cover.

Astrophotography is a continual learning experience; it is all too easy to plough in and follow established, familiar processing workflows. Hindsight is a wonderful thing too. Having calibrated, registered and integrated the separate channels, it was questionable that the time spent acquiring luminance may have been better spent on Hα exposures. If one compares the two grey-scale images in fig.1, it is clear that the Hα image has better red nebulosity definition and a tighter star appearance too. Nebulosity of other colors though is more muted but Alnitak has a much improved appearance.

The binned RGB color exposures show some blooming on bright stars and they require extra care during the registration process to avoid ringing artefacts. The outcome repeats some of the lessons from the Dumbbell Nebula example, where the wideband luminance information was similarly challenged in favor of narrowband luminance data. The key takeaway is to expose an image through each filter, apply a screen stretch and examine each carefully for its likely contribution to the final image before deciding on the final image acquisition plan. That is easier said than done, since it is difficult to predict how subsequent processing will play out, but it is better than following a standard acquisition plan that completely disregards the contribution of each channel to the final image.

fig.1 The image on the left is a processed luminance stack taken through a clear filter. The image on the right was taken with a Hα filter. In the presence of mild light pollution the Hα image is superior in almost every way for defining stars and nebulosity.

Linear Processing

After checking the individual L,RGB and Hα frames for duds, they were calibrated and registered. In the case of the binned RGB exposures, the standard registration parameters using an Auto interpolation method typically default to Lanczos 3 and a Clamping threshold of 0.3, which produces ringing artefacts. For these channels, I switched the interpolation method to Bicubic Spline and reduced the clamping threshold of 0.1. The lower clamping threshold reduced ringing artefacts but degrades fine detail. This was not an issue as the RGB data was principally used for low-resolution color information. Image integration followed, using Winsorized Sigma algorithms and Sigma thresholds set to just remove cosmic ray hits and various spurious trails for satellites, meteorites and aircraft. The individual image stacks, while in their linear state, had MureDenoise applied to them, with the settings adjusted to the individual interpolation method, gain and noise for each binning level and image count.

fig.2 The LRGB image above uses the processed luminance channel taken through a clear filter. The larger star sizes hide the blue halos in all but the largest stars.

Luminance Strategy

As mentioned earlier, the screen-stretched Luminance and Hα images in fig.1 play out an interesting story. The light pollution clearly washes out the faint red nebulosity in the broadband luminance image and at the same time, the brightest stars are bloated and ugly. The Hα channel looks considerably better but using it for luminance data causes subtle issues around bright stars later on. Using Hα for luminance naturally favors its own and often leads to a red-dominated picture. Although red nebulosity is abundant in this image, it gives the flame nebula a red hue too and in addition, diminishes the appearance of several blue reflection nebula close to the familiar horse head.

fig.3 In this HαRGB image, the smaller stars are surrounded by blue halos (these are not deconvolution artefacts and are not present in the luminance channel).

The image in the first edition used a blend of luminance with a little Hα, to form a master luminance file. This image has obvious diffraction artefacts around Alnitak and the other bright stars. The next attempt used the Hα channel for luminance. At first glance, the stretched Hα channel looked very promising, not particularly surprising considering the overall red color of the nebula, although it lacks some definition in NGC2023. Further down the road, however, there is a bigger issue when it is combined with the RGB image: The star sizes in the images taken with narrowband filters are considerably smaller than those through broadband filters and, after applying deconvolution and masked stretch, become smaller still. This is normally good news but after LRGB combination, many small bright stars are surrounded by a blue halo (fig.3).

This highlights a tricky issue and makes this image an interesting case study. It is easy to “plug in” stars against a dark background but in order to do so over a lighter background requires a close structure match between the luminance and color data to yield a natural appearance. In this image many stars are surrounded by bright red nebulosity and hence high luminance values. In the RGB file, these areas are dominated by diffuse blue reflections around the brighter stars. This causes the blue coloration around the brightest stars, overriding the red background. The range of issues experienced with the LRGB blends is shown in fig.2 and fig.3. One bookend using Hα for luminance has tight stars with blue halos, the other uses standard luminance exposures has larger stars and diffraction artefacts. In a perfect world, we need to retain the star information from the Hα channel but minimize the blue halos around stars. This requires some subtle trade-offs; several alternative strategies come to mind, including:

• experiment with different combinations of Luminance and Hα to create luminance channel

• shrink the stars (and halation) in the RGB channels

• histogram stretch the Hα channel (rather than MaskedStretch) to avoid star erosion

• separately process the RGB images for small (star) structures and larger general coloration

There is seldom a silver bullet for these kinds of issues and all the above strategies were tried in various combinations. The outcome is an optimized compromise. As channel blending and linear fit algorithms are confused by general background levels and the background of broadband and narrowband images are very different, the background levels were subtracted by applying the DynamicBackgroundExtraction tool to each integrated stack. A clumsy use of DBE also removes faint nebulosity and hence the background samples were placed very carefully to avoid any such areas. A “superlum” image was built up by first combining the short and long luminance exposures using the HDRComposition tool. This blends the clipped stars in the 5-minute exposures with the unsaturated pixels from the shorter exposures. The final “superlum” file was created by combining L and Hα files, allowing their contributions to be scaled according to their noise levels. This produces the lowest noise level for the general image background and a compromise between star sizes and artefacts. Deconvolution was carefully set to reduce star sizes a little and give better definition to the brighter nebulosity (fig.4). (In this study, the RGB exposures were binned and showed some blooming on the brightest stars. Since registration scales these images it is technically possible to integrate with the 1×1 binned files. The result is not pleasing and in this instance I did not use them to contribute to the “superlum”. I now take unbinned RGB exposures for maximum flexibility during processing.)

fig.4 Deconvolution on this image requires progressive deringing settings for stars in dark regions, local support for stars in bright nebulosity and regularization to avoid artefacts in bright regions. A range mask also protected dark regions.

Non-Linear Processing

Both luminance and color images were stretched in the same manner to maintain, as much as possible, the same star sizes. This comprised of an initial, medium HistogramTransformation, which boosted the brighter stars and produced a faint image, followed by Masked-Stretch. This departure from an all HT non-linear stretch, prevented the brighter stars clipping and kept their size under control.

The bright star sizes in the color image were reduced further, using a star mask and the MorphologicalTransformation tool. The StarMask noise threshold was set to detect the brightest stars, with sufficient structure growth to encompass their immediate surroundings. Some convolution was applied to the star mask to soften and extend the boundaries. This also had the effect to draw in the red surroundings around the medium bright stars to replace their blue halos. In the case of the few super-bright stars, the extent of their blue halo was too large to remove and remains in the image. Not ideal I admit, but a quick Internet image search suggests I’m not alone! As usual, star color was enhanced using the color saturation tool in conjunction with a star mask. The tool was set to enhance the yellow and red stars and to decrease the blue color saturation (fig.6). In preparation for LRGBCombination, a small amount of convolution was applied to the image (with the star mask in place) followed by a healthy dose of noise reduction, using TGVDenoise, to remove chromatic noise.

The stretched “superlum” channel had its structures enhanced and sharpened with successive applications of LocalHistogramEqualization (LHE) at different large scales and sharpened with MultiscaleMedianTransform (fig.5). In between applications, the dynamic range was extended slightly (to avoid clipped highlights) by applying a plain HistogramTransformation with a 10% highlight extension. After tuning the brightness distribution with a S-curve CurvesTransformation, I applied selective noise reduction, using MultiscaleLinearTransform and replaced green pixels with neutral ones, using Selective-ColorNoiseReduction (SCNR).

As usual, the color and luminance channels are combined using the LRGBCombination tool. Before using this tool, however, I followed a suggestion from the PixInsight forum. In this process, the luminance values in the RGB and master luminance are equalized before the application of the LRGBCombination tool. To do this, I first decomposed the RGB file using the ChannelExtraction tool set to CIE L*a*b* mode. I then applied the LinearFit tool to the L channel, using the “superlum” as the reference image. After using the ChannelCombination tool to put them back together I followed up with the LRGBCombination tool as normal. After doing this extra step, the result is more predictable and easier to tune with small slider changes.

In the final image, slightly larger star sizes are the sacrifice for fewer small blue haloes. The MaskedStretch certainly reduces the visual dominance of the brightest stars and the blending process records both the blue and red nebulosity. I noticed, however, that the dominance of the Hα signal in both the luminance and color files caused the flame nebula to turn pink. In one-shot color images, this is more orange in appearance and for the final image, the color was gently adjusted in Photo-shop using a soft-edged lasso and the hue tools. I did contemplate using the spherical blur tool on Alnitak to remove the diffraction artefacts but ultimately resisted the temptation; if it were that easy to remove the effects of subtle optical anomalies, I would not have an excuse to upgrade my refractor.

fig.5 After gentle boosts to nebulosity from repeated applications of the LocalHistogramEqualization tool (at different large scales) I used MMT to sharpen up smaller structures. A linear mask protected dark areas.

fig.6 Star color saturation was selectively adjusted with the ColorSaturation tool (blue/green saturation is reduced).

fig.7 This is the simplified workflow of the separate color, luminance and narrowband channels and combinations. By now, you will appreciate that the difference between good and also-ran images lies in the subtlety of the various tool settings as well as the tool selection and order. Patient experimentation is the key ingredient for many deep sky images. Frequent breaks during image processing help calibrate perception. Returning to an image helps overcome the tendency to over-process.