Histogram Stretching

Presently we have at least two image files, the monochrome luminance file and an RGB color file, and we stretch these non-linearly and independently before combining them. Non-linear stretching involves some form of tonal distortion. Changing the range of a file, via an endpoint adjustment, still results in a linear file. In PixInsight, stretching is achieved most conveniently with the HistogramTransformation tool (fig.1). In essence the mid tones slider is dragged to the extreme left-hand side creating a large increase in shadow contrast. It is important to not change the right-hand highlight slider. This ensures the transformation does not clip bright pixels. Drag the left-hand slider, which controls the black point very carefully to the extreme left-hand side of the image histogram, just before the readout shows clipped pixels. (This is shown along with the percentage of clipped pixels in the tool dialog box.) Putting aside clipped pixels, this manipulation is a purely subjective adjustment, judged from the screen appearance. For that reason it is essential to disable the screen transfer function (STF) to accurately assess the effect. The HistogramTransformation tool is one of those that has a real-time preview to assess the adjustment. It is also easier to obtain the desired effect by making the image stretch in two passes, with the first making about 80% of the total adjustment. Now reset the histogram tool and perform a second lesser stretch with greater finesse and finer control. To precisely place the sliders, home into the shadow foot of the histogram with the toolbox’s zoom buttons. Repeat this exercise for both the color image and the luminance image, noting that they may need a different level of adjustment to obtain the desired result. Like many other users you may find the screen transfer function automatically provides a very good approximation to the required image stretch. A simple trick is to copy its settings across to the histogram transfer tool, by simply dragging the blue triangle from the STF dialog across to the bottom of the histogram dialog box. Voilà!

This is just one way to perform a non-linear stretch in one application. If you prefer, one can achieve similar results using repeated curve adjustments in Photoshop, Nebulosity, Astroart or Maxim DL. Maxim DL also has a histogram remapping tool which you may find useful. In any every case, one thing is essential; the image must have 16-bit or preferably 32-bit depth to survive extreme stretching without tone separation (posterization). PI defaults to 32-bit floating point images during the stacking and integration process. If an image is exported to another application, it may require conversion to 16-bit or 32-bit integer files for compatibility.

When I started astrophotography I tried doing everything on my existing Mac OSX platform. My early images were processed in Nebulosity and I used the Curves tool, DDP and the Levels / Power Stretch to tease the detail out of my first images. To achieve the necessary stretch, the Bezier Curves and the Power Stretch tool were repeatedly applied to build up the result. Without masks to protect the background and stars, many of my early images featured big white stars and noisy backgrounds, which then required extensive work to make then acceptable.

Digital Development Process

No discussion on stretching can ignore this tool. DDP is an interesting dilemma. It is a combination of a sharpening and stretch tool originally designed to mimic the appearance and sensitivity of photographic film. When applied to a linear image, it gives an instant result, and for many, the allure of this initially agreeable image dissuades the astronomer from further experimentation. There are versions of this tool in Nebulosity, Maxim and PixInsight to name a few. It is very useful to quickly evaluate an im-age’s potential. The results, though, are very sensitive to the background threshold and sharpening settings and the result often lacks refinement. This includes excessive background noise and star bloat. The sharpening algorithms that sometimes accompany this tool are also prone to produce black halos around bright stars. With a little patience and experimentation superior results are obtained by more sensitive methods. It is interesting to note that PixInsight have relegated their implementation of the tool to its obsolete section.

fig.2 The basic building blocks of the non-linear workflow for an LRGB image. It follows on from the workflow in the previous chapter on Linear Processing and assumes separate luminance and color exposures. This general process applies equally to many applications, with a little variation.

fig.3 Maxim DL also has an option to use the screen stretch settings to set the endpoints on a logarithmic or non-linear gamma curve. This has produced a good initial stretch to the entire image, in this instance, on the luminance file.

Masked Stretching

Some images do not take kindly to a global image stretch. In particular, those with faint nebulosity and bright stars are particularly troublesome. The same manipulation that boosts the faint nebulosity emphasizes the diffuse bright star glow and increases their size. Any form of image stretching will enlarge bright stars and their diffuse halo to some extent. It is an aesthetic choice but most agree that excessive star size detracts from an image. Some star growth is perfectly acceptable but if left unchecked, it alters the visual balance between stars and the intended deep sky object. Some workflows tackle star size later on in the imaging process with a transform tool, for example the MorphologicalTransformation tool in PI or, one at a time, using the Spherize filter in Photoshop. An alternative method preferred by some is to stretch a starless image, either by masking the stars before applying the stretching tool or by removing them altogether. These advanced techniques take some practice and early attempts will likely produce strange artefacts around bright stars.

Not all applications have the ability to selectively stretch an image. Maxim DL 5 (fig.3) and Nebulosity presently only support global application and for instance would require an external manipulation to remove stars from the image prior to stretching. PixInsight and Adobe processes support selective application with masks. The latest version of PI introduces the new MaskedStretch tool. In our example image, it produces artefacts on bright stars, regardless of setting, and is more effective, however, when used on a partially stretched luminance file. In this case, apply a medium strength HistogramTransformation curve to the luminance channel and then follow with a masked stretch. Altering the balance between the two stretches gives a range of effects. Applied to our ongoing example gives better definition in faint details and less star bloat than the same luminance file simply processed using the HistogramTransformation tool in combination with a star mask. The results are subtly different in other ways too: I prefer the balance between the nebulosity and the star field in the luminance file with masked stretching. Initial trials show an improvement in the nebulosity definition at the expense of a little noise.

fig.4 Maxim DL, in common with other astro imaging programs, has a digital development process tool. This stretches the image, with options for smoothing and sharpening. In the case of Nebulosity 3 and Maxim DL 5, there is no mask support.

fig.5 A combination of a medium HistogramTransformation and a MaskedStretch produces a pleasing balance of detail in the nebulosity and with less star bloat. In the two previews above, the result of a standard stretch is on the left and the masked stretch on the right.

Star Removal

One extreme approach is to remove the stars from the image, stretch what remains and then put the stars back. This is particularly useful with images that are exposed through narrowband filters. The extreme stretches involved to tease details out of SII and OIII wavelengths create ugly stars at the same time, often with unusual colors. In fact, some practitioners permanently remove stars from their narrowband files and, in addition to the many hours of narrow band exposures, they expose a few RGB files specifically for the star content. The RGB files are processed to optimize star shape and color and the stars are extracted from the file and added to the rainbow colors of the processed narrow band image.

To remove the stars temporarily there are several means to erode stars until they disappear. In Photoshop this is the dust and scratches filter, applied globally. Its original purpose is to fill in small anomalies with an average of the surrounding pixels. Luckily for us it treats stars as “dust” and in simplistic terms, if used iteratively, with different radii, diminishes star size until they disappear. The process is a little more involved than that and some process workflows have dozens of individual steps. This is a prime candidate for a Photoshop action and indeed, J-P Metsävainio at www.astroanarchy.blogspot.de has a sophisticated Photoshop plugin to do just that (donate ware). Another Photoshop method is to try the minimum filter. After creating a star mask using the Color Range’s dropper on progressively dimmer stars, expand the selection by one pixel and feather by two. To finish off, apply the Minimum Filter with its radius set to 1 pixel.

In PI, the MorphologicalTransformation tool mentioned earlier has a lesser effect than the dust and scratches tool, even after repeated application. A third alternative is a low cost utility called Straton (fig.7), available from the website www.zipproth.com. This is very effective and convenient to use. With the stars removed, a star-only image is created from the difference of the original image and the starless one. With two images, stretching and enhancement is easily tailored to the subject matter in each case, with less need for finessed masking. Once completed, the two images are added back together. Combining images is not that difficult: To add or subtract images, simply select the linear dodge or subtract blending modes between Photoshop layers containing the star and starless images. Alternatively, image combination is easily achieved with the PixelMath tool in PI or Maxim DL either to combine the two images or generate their difference. In this case a simple equation adds or subtracts pixel values from each other to create a third image.

Photoshop Techniques for Increasing Color Saturation

Photoshop is at home with color images and has many interesting methods to improve color saturation. They make use of the color blending mode and layers to enhance image color. Many of these techniques follow the simple principle of blending a slightly blurred boosted color image with itself, using the color blending mode. The blurring reduces the risk of introducing color noise into the underlying image. The differences between the methods lie in the method to achieve the boosted color image.

One subtle solution is to use a property of the soft light blending mode. When two identical images are combined with a soft light blending mode, the image takes on a higher apparent contrast, similar to the effect of applying a traditional S-curve. At the same time as the contrast change, the colors intensify. If the result of this intensified color is gently blurred and combined with another copy of the image using the color blend mode, the result is an image with more intense colors but the same tonal distribution. In practice, duplicate the image twice, change the top layer blending mode to soft light and merge with the middle layer. Slightly blur this new colorful and contrasty image to deter color noise and change its blending mode to color. Merge this with the background image to give a gentle boost to color saturation and repeat the process to intensify the effect. Alternatively, for an intense color boost, substitute the soft light blending mode used on the top layer for the color burn blending mode. If that result is little too intense, lower the opacity of the color layer before merging the layers.

The other variations use the same trick of blending a slightly blurred boosted color image with itself. At its simplest, duplicate the image, apply the vibrance tool (introduced into Photoshop several years ago), blur and blend as before using the color mode. Another more involved technique, referred to as blurred saturation layering, works on separate luminance and color files using the saturation tool. The color saturation tool on its own can be a bit heavy-handed and will introduce color noise. This technique slightly lowers the luminance of the RGB file before increasing its saturation and then combining it with an unadulterated luminance file. In practice, with the RGB file as the background image, create two new layers with the luminance image file. Set the blending mode of each monochrome image to luminosity, blur the middle layer by about a pixel and reduce its opacity to 40%. Merge this layer with the RGB layer beneath it and boost its color saturation to taste.

Both PixInsight and Photoshop exploit the LAB color mode and its unique way of breaking an image into luminosity and two color difference channels. Simply put, if the equal and symmetrical S-curves are applied to the two color difference channels a and b, the effect is to amplify the color difference and keep the overall color balance. Pix-Insight flits between color modes and uses this technique behind the scenes to increase color saturation. Photoshop users have to use a procedure: In practice, convert the image mode to LAB, duplicate it and select the top layer. As before, change the blending mode to color and apply a small Gaussian blur to it. Open its channel dialog and select the a channel. Apply a gentle s-curve adjustment to it. (Use the readout box in the Curves dialog to make equal and opposite adjustments on each side of the center point.) Save the curve as a preset and apply it to the b channel as well. As before, the trick is to blend a slightly blurred saturated image, using the color blend mode, with the original image. With the LAB technique, the steeper the curve as it crosses the middle peak of the a and b histograms, the greater the color differentiation.

Noise Reduction

The example image suffers from excessive bias pattern noise, most noticeable in areas of faint nebulosity, even after calibration. In PixInsight, the ACDNR tool is an effective tool for reducing noise in non-linear images, as is the MultiscaleLinearTransform tool, both of which can work at a defined structure size. (PixInsight is a dynamic program with constant enhancements and developments; a year ago the ATrousWaveletTransform tool would have been the tool of choice.) In this example, the sensor pattern noise occurs between alternating pixel columns, that is, the smallest scale. It is selected with the structure size set to 1 and the noise reduced by averaging the values over a 3x3 pixel area with a weighted average. With many noise reduction algorithms, little and often is better than a single coarse setting. This tool has a dual purpose: It can emphasize structure and reduce noise at the same spatial scale. With so many choices, the real-time preview comes into its own and is invaluable when you experiment with different settings and scales to see how the image reacts. In the case of ACDNR tool, there are further options: It can remove luminance noise (the lightness tab) or color noise (chrominance tab). It also has settings that prevent star edges from being blurred into the background (Bright Sides Edge Protection).

A good starting point for selective noise reduction, say with the MultiscaleLinearTransform (MLT) tool, is to progressively reduce the noise level at each scale. One suggestion is to set noise reduction levels of 3, 2, 1 and 0.5 for scales 1 to 4 and experiment with reducing the noise reduction amount and increasing the number of iterations at each scale.

fig.13 A repeat of fig.9, the MLT tool can be used for noise reduction and for emphasizing structures at the same time. Here a combination of mild sharpening and a boost of medium size structures emphasize the nebulosity structure and sharpens up the stars at the same time. The first three layers have a decreasing level of noise reduction. Layers 2, 3 and 4 have a subtle boost to their structures. The tool also has options for deringing and noise sensitive settings that discriminate between good and bad signal differences. For even more control, it can be applied to an image in combination with a mask, to prevent unwanted side effects.

fig.14 The HDRMultiscaleTransform can also be applied to the LRGB image to enhance structure. In this case, it has been applied to the luminance information in the combined LRGB image with deringing and a lightness mask enabled to avoid unwanted artefacts.

Finally, if green pixels have crept back into the image, apply the SCNR tool to fix the issue, or the Photoshop process described in the last chapter, by using the color range tool and a curve adjustment.

Sharpening

Noise reduction and sharpening go hand in hand, and in PixInsight, they are increasingly applied in combination within the same tool. The MLT tool can not only be used for progressive noise reduction at different scales, but also to boost local contrast at a scale level, giving the impression of image sharpening. As with other multi-scale tools, each successive scale is double the last, that is 1, 2, 4, 8 pixels and so on. (There is also a residual scale, R, that selects the remaining items that make up the large scale structures within the image.) In the MLT tool, the bias and noise reduction level is individually set at each image scale. When the bias is set to zero, the structures at that scale are not emphasized. The noise settings on the other hand potentially blur structures at that scale and de-emphasize. There is a real-time preview to quickly evaluate the effect of a setting. I often prefer to apply it to a small preview at 50% zoom level and directly compare it with a clone preview. I have no doubt you will quickly discover the undo / redo preview button and its shortcut (cmd-Z or ctrl-Z, depending on platform).

A second tool that is used to create the impression of sharpening is the HDRMultiscaleTransform tool. This is applied either to the luminance file or to the luminance information in the LRGB image. In the running example it was applied to the luminance image (fig.8) and by way of comparison to the LRGB image (fig.14).

Enhancing Structures with Photoshop

Whilst on the subject of local contrast enhancement, Photoshop users have a number of unique tools at their disposal, the first of which is the HDR Toning tool. This tool works on 32-bit images and is very effective. Indeed, it is a surprise to find out it was not designed for astro-photography in the first place. It is one of those things that you discover through patient experimentation and exploit for a purpose it was not designed for.

Real High Dynamic Range processing combines multiple exposures of different lengths into a single file. To compress all the tones into one image requires some extreme tonal manipulation. The obvious method simply compresses the image tonally between the endpoints of the extreme exposures. That produces a very dull result. More usefully, the HDR Toning tool in Photoshop can apply local histogram transformations to different areas of the image to increase contrast and then blend these together. On conventional images this creates a peculiar ghostly appearance that is not to everyone’s taste. In astrophotography, it is particularly effective since the groups of image elements are often separated by a swathe of dark sky that disguises the manipulation. Just as with the multi-scale operations in PixInsight, local contrast is applied at a certain user-selectable scale. In the example in fig.17 and fig.18, the scale was selected to emphasize the detail with the bubble and outer clouds. The vibrance setting increases color saturation without clipping.

The second Photoshop technique used for enhancing structures uses the high pass filter and layer blending modes. This method is able to emphasize contrast at different scales in a broadly similar manner to the multi-scale processing in PixInsight. At its heart is the high pass filter. When applied to an image on its own, it produces a very unpromising grey image. When you look in closer, you can just make out the boundaries of objects, picked out in paler and darker shades of grey. The radius setting in the high pass filter dialog determines what is picked out. The “a-ha” moment comes when you blend it with the original image using the overlay blending mode. Normality is resumed but the image structures are now generally crisper, with enhanced local contrast where there were faint lines of pale and darker grey in the high pass filter output. This enhancement is normally applied in conjunction with a luminosity mask, to protect the background. Just as with multi-scale processing, it is sometimes necessary to repeat the treatment at different pixel radii to show up multiple structures.

fig.17 When you first open the HDR Toning tool with a 32-bit image, set the method to Local Adaptation and reset the tone, detail and advanced sliders to their mid positions and the edge glow settings at a minimum. The tone curve should be a straight line. The tool settings interact and at first, just play with small changes to get a feel for the effects. The edge glow settings are particularly sensitive and define what image scales benefit from the contrast enhancement.

fig.18 After 10 minutes of experimentation a combination of a subtle tone curve, increased color vibrancy and increased detail make the image pop. A subtle amount of edge glow gives the nebula’s cloud fronts a 3D quality. In this example, the histogram starts at the midpoint as the TIFF output from PixInsight converts into an unsigned 32-bit integer.

One-Shot Color (OSC) / RGB

In the case where the image is formed of RGB data only, either from a sensor sandwiched to a Bayer filter (i.e. any conventional digital camera) or through separate RGB filters, there is no separate luminance channel to process. The same guidelines apply though; it is the luminance data that requires sharpening and deconvolution. In this case, extract the luminance information from the stacked RGB file and treat as a separate luminance file through the remainder of the process. (Each of the astro imaging programs have a dedicated tool for this and if you are a Photoshop user, you can accomplish the same by changing the image mode to Lab and extracting the luminance (L) channel from the channels dialog.)

Compared to straightforward RGB processing, image quality improves when a synthetic luminance channel is extracted, processed and combined later on. This is good advice for a color camera user. Those who use a filter wheel might be wondering, why not shoot luminance exposures too? After all, many texts confirm that it is only necessary to take full definition luminance files and that lower resolution color information is sufficient and the binning gives the added bonus of shorter exposures / improved signal to noise ratio. There is a school of thought, however, that believes that LRGB imaging (where the RGB data is binned 2x2) in a typical country town environment with some light pollution, may be improved upon by using RGB imaging alone and without binning. Unlike the RGB Bayer array on a color camera, separate RGB filter sets for filter wheels are designed to exclude the principal light pollution wavelengths. The yellow sodium emission wavelength falls neatly between the red and green filter responses. The argument proposes that for any given overall imaging time, a better result is obtained by using RGB at 1x1 binning than L (1x1) and RGB (2x2). It reasons that the exposures through the luminance filter include a good deal of light pollution and its associated shot noise. This shot noise exceeds that of the shot noise associated with faint deep sky objects. Over the same overall imaging time, the separate RGB exposures have a better combined signal to noise ratio and if binned 1x1, have the spatial information to provide a detailed image. At the same time, their narrower bandwidth is less likely to clip highlights on bright stars, as so frequently occurs in luminance exposures.

fig.19 The high pass filter in Photoshop can act as a multi-scale enhancement tool. From the top, the original file. Next, the background layer is duplicated twice and the high pass filter (filter>other>high pass) with a radius setting of 4 is applied to the top layer. The blending mode is set to overlay and the whole image sharpens up. In the third box, the two top layers are merged and a luminance layer mask is added. To do this, the background image is copied (cmd-A, cmd-C) and the mask is selected with alt-click. This image is pasted in with cmd-v. (These are Mac OSX keyboard shortcuts. The Windows shortcuts usually use ctrl instead of cmd.) This mask image is slightly stretched to ensure the background area is protected by black in the mask. Clicking on the image shows the final result. In the final image, the stars and nebula details are both sharpened to some degree. In practice, some use a star mask to protect the stars from being affected or remove the stars altogether before applying this technique.

I have tried both approaches from my back yard and believe that separate RGB images processed with synthetic luminance (channel combination weighted by their noise level) certainly give excellent results with rich star fields and clusters. The star definition and color was excellent. This interesting idea requires more research and two successive clear nights for a proper back to back comparison. This is not a frequent occurrence in the UK!

Image Repairs

If things have gone well, image repairs should not be necessary. If some problems remain Photoshop is in its element with its extensive cosmetic correction tools. In the latest versions it has intelligent cloning tools that replace a selection with chameleon-like camouflage. One such tool uses the new “content aware” option in the fill tool. The problem area is simply lassoed and filled, ticking the content aware box. The result is remarkable. A similar tool, particularly useful for small round blemishes, is the spot healing brush. In practice, select a brush radius to match the problem area and select the “proximity match” option before clicking on the problem. These tools were originally designed for fixing blemishes, particularly on portraits. As the T-shirt slogan says, “Photoshop, helping the ugly since 1988”!

fig.20 If there is no separate luminance exposure, the trick is to create one. After processing the color information, up to the point of non-linear stretching, extract the luminance information and process it separately as a luminance file as in fig.2. (For users of one-shot color cameras, extract the luminance information just before non-linear stretching.) The quality improvement is substantial over RGB-only processing.

Correcting Elongated Stars

In addition to the MorphologicalTransformation tool in PixInsight (one of its options identifies and distorts stars back into shape) Photoshop users have a few options of their own to correct slight star elongation. If the image is just a star field, you may find the following is sufficient: Duplicate the image into a new layer, set its blending mode to darken and move the image 1 pixel at a time. This will only work up to a few pixels and may create unwanted artefacts in galaxies or nebulosity. Another similar technique is more selective: In the “pixel offset technique”, rotate the image so the elongation is parallel to one axis, duplicate it into another layer and select the darken blend mode. Using the color range tool, select bright stars in the duplicated layer and add to the selection, until most of the stars are identified. Modify the selection by enlarging by 1 or 2 pixels and feather by a few pixels to create a soft edged selection of all the stars. Now choose the offset filter (filter>other>offset) and nudge by 1 or 2 pixels. Once the desired effect is achieved, flatten the layers and de-rotate.

Big oblong stars pose a unique problem. One way to fix these is to individually blur them into a circle with the radial blur tool (filter>blur>radial blur)and then reduce their size with the spherize filter (filter>distort>spherize). In Adobe CS6, the image has to be in 8-bit mode for this tool to become available and for that reason, this cosmetic fix should be one of the last operations on an image. This tool will likely distort neighboring stars or move them. If you duplicate the image and apply the filter to the duplicate, you can paste the offending stars back into their correct positions from the background image.

Correcting Colored Star Fringing

Even though the RGB frames are matched and registered, color fringes may occur on stars as a result of focusing issues and small amounts of chromatic distortion. This is a common occurrence on narrowband images too, caused by the significant differences in stretching required to balance each channel. Photo-shop users have a few tools with which to tackle these, depending on the rest of the image content.

fig.21 Photoshop has several cosmetic defect tools. Here is an evaluation of a content-aware fill and a spot healing brush set to “proximity match”. The grey halo around a prominent star in the middle of the bubble is selected by two circular marquees and feathered by a few pixels. The image on the right shows the result of a content aware fill and the one on the left, the spot healing brush. Note the spot healing brush has also pasted in several small stars that can be seen at the 10 o’clock position.

If the color of the fringe is unique, one can select it with the color range tool and neutralize it by adjusting the selection by adding the opposing color. If the stars are mixed up with nebulosity of the same color, this technique will also drain the color from the nebulosity. In this case, a star mask may not work, as each star may end up with a small grey halo around it. An alternate solution is to try the chromatic aberration tools in Photoshop (filter>lens correction>custom>chromatic aberration) and adjust the sliders to remove the offending color. Be careful not to go too far, or it will actually introduce fringes.

Extending Faint Nebulosity

When an image has an extended faint signal it is useful to boost this without affecting the brighter elements of the image. The PI tool of choice for emphasizing faint details is the LocalHistogramEqualization process. If set to a large radius, it will emphasize the contrast between large structures rather than at a pixel level and emphasize noise. The trick is to apply the LHE to the image through a mask that excludes stars and brighter areas of the image. This is accomplished by a compound mask, made up of a star mask and one that excludes bright values. The combination of these two images breaks the ice with the PixelMath tool (figs.22, 23). In the ongoing example of the Bubble Nebula, we first check the settings in a small preview and then apply to the full image to see the overall effect (fig.24).

In the first instance, we use the StarMask tool to make a normal star mask. It should be distinct and tight; there is no need to grow the selection by much and select a moderate smoothness. If it is set too high, especially in a dense star field, the soften edges of the mask join up and there is too much protection of the intervening dark sky. Set the scale to ensure the largest stars are included. Having done that and checked that the resulting mask file, minimize it for later use. Now open the RangeSelection tool and click on the real-time preview. Increase the lower limit until the brighter areas of nebulosity show up in white. Apply a little smoothness to remove the hard edges and apply this tool to the image. We now have two separate masks and these are combined with PixelMath. You can see in fig.23 that combining the images in this case is a simple sum (or max) of the two images. Ensure the output is not re-scaled so the result combines both masks and clips them to black and white. (If the output was re-scaled, the mask would be tri-tone, white, black and grey.)

This mask is now applied to the image and inverted, to protect the light areas. With it in place, the subtle red nebulosity is boosted with the LocalHistogramEqualization tool, with a kernel radius set around 100–300 (to emphasize large cloud structures) and the contrast limit between 1 and 3. After checking the settings in a preview window, it is applied to the main image. Side by side, the effect is subtle and may not read well off the printed page. In this example, the faint wispy areas of nebulosity are more prominent in relation to the dark sky and the image has an overall less-processed look. This particular example uses a color image but it is equally effective when applied to the luminance file.

fig.22 The StarMask and RangeSelection tools are adjusted to select stars and bright areas of nebulosity. The critical area is the vicinity of the bubble and the preview is used to quickly check the mask extent.

fig.23 The two masks generated in fig.22 are combined using PixelMath. Click on the Expression Editor and select the filenames from the drop down list. Open up the Destination settings and check create new file and deselect re-scale output. Apply this mask to the image by dragging its tab to the image left hand border. Invert the mask to protect the highlights.

fig.24 With the mask in place, check the LHE tool settings using the real-time preview and then apply to the image. In this example, the original image is on the left for comparison. The right hand image shows lighter wispy detail in the star field.

fig.25 The best time to remove bad pixels is during image calibration. Sometimes a few slip through and may be obtrusive in lighter areas. The two tools opposite will detect a dark pixel and replace it with a lighter value. The CosmeticCorrection tool has a simple slider that sets the threshold limit for cold pixels. The real-time preview identifies which pixels will be filled in. Alternatively, PixelMath can do the same thing with a simple equation and give a little more control. Here, if a pixel is lower than 0.1, it is replaced with a blend of the pixel value and the median pixel value of the entire image (which is typically a similar value to the average background value). A little experimentation on a preview window determines the detection threshold and the degree of blending. If a dark pixel has spread, try slightly under-correcting the problem but repeat with two passes and with a slightly higher threshold on the second pass.

Removing Dark Pixels

Image calibration sometimes introduces black pixels into an image, or they occur with later image manipulation that creates an increase in local contrast. Even without dark frame auto-scaling, the image calibration in PixInsight or Maxim may over-compensate and conflict with the camera’s own dark processing. This introduces random dark pixels. While these are not noticeable in the background, they detract from the image when they occur in brighter areas or after stretching. Isolated black pixels also seem to resist noise-reduction algorithms. The solution is to replace these cold pixels with an average of their surroundings. The CosmeticCorrection tool has the ability to detect cold and hot pixels and has the convenience of generating a preview of the defect map. Dragging the Cold Sigma slider to the left increases the selection threshold and the number of pixels. These pixels are replaced with a blend of surrounding pixels.

An alternative is a simple conditional statement using PixInsight’s PixelMath tool. This selects pixels with a value lower than a defined threshold and substitutes them with an average background value. The sensitivity is determined by the threshold value in the equation. In this case it is 0.1. Both of these have no blending effect and so they literally substitute pixel values. For this reason, defects are best removed before the cold pixel boundary has blurred into neighboring pixels, or the fixed pixels may retain a small dark halo. Alternative blending equations can be used to combine the current pixel value with another. The tool can also be applied iteratively to great effect.