| Home | Galleries | Articles | Reviews | Best Gear | Science | New | About | Contact |
by Roger N. Clark
Astrophotography post processing with images made under moderate light pollution can be done simply with modern raw converters and image editors that can edit in at least 16-bits/channel.
The Night Photography Series:
Contents
Introduction
Raw Conversion
Image Stacking
Light Pollution and Airglow Removal
Enhance
Star Diameter Reduction
Is the Color Gradient Correct?
Try it Yourself With These Images
About the Challenge
Diagnosing Problems Using Histograms
White Balance and Histogram Equalization
Conclusions
References and Further Reading
Below I outline the steps needed to make beautiful astrophotos made under moderate light pollution skies. A series of nine 1-minute exposures of the Scorpio region and the Rho Ophiuchus nebula complex was made with a 100 mm lens at f/2 and a stock Canon 6D, tracking on the stars. Figure 1a shows my finished image. One of the nine images, an out of camera jpeg, is shown in Figure 1b. As you can see from Figure 1b, the light pollution and airglow has washed out the image. Multiple exposures, when combined, increases the signal-to-noise ratio, allowing for weaker signals to be extracted. The signal-to-noise ratio increases with the square root of the number of exposures, so the 9 exposures in this example improves the signal-to-noise ratio by 3 times.
The sky images were obtained from a "green zone" (on the light pollution scale) west of Denver at an elevation of about 11,000 feet with no Moon. There is a light pollution gradient from left to right, as well as airglow. The Milky Way is on the left edge, so there is also a deep sky brightness gradient from the galaxy. Thus all together, with the relatively short total exposure time, it will be a challenge to bring out the nebula, including dark nebulae, hydrogen alpha nebulae, blue and the yellow reflection nebulae.
The challenge is to produce the best image that brings out the many colorful nebulae in the region, by removing the veil of the light pollution and airglow. Examine Figure 1c.
Details of the histogram from Figure 1b are shown in Figure 1c. We know many parts of the night sky are extremely dim, so the large offset, labeled "Light Pollution + Aiglow" is light added to the signal from the deep sky beyond Earth, and that the offset is due to light from cities and other artificial light scattered in the Earth's atmosphere plus light emitted by the atmosphere (airglow). Note that the offset is different for each color. That means a different offset needs to be subtracted for each color. The middle section of the histogram, where each color peaks, is where the light from nebulae, galaxies, fainter stars and other deep sky objects is represented, mixed in with gradients in light pollution and airglow. Airglow often shows as bands, sometimes bands of red an green. Thus, besides the offset shown by the left side of the histogram, there are mixed signals in the histogram peak that need to be separated, and the weak signals from the night sky beyond the Earth's atmosphere enhanced so we may see them (this is called image stretching).
I'll show below, that a single constant offset can be done in the raw converter (Figure 2, right-most panel, blacks set to -83 in this case). Then show how to subtract the different constants for each color, then how to subtract gradients and enhance the signals from the night sky to bring out all the wonderful nebulae and their various colors.
The basics of the effects of the curves tool are shown in Figure 1d. Light pollution should be subtracted, and you can do this by moving the lower left point in the curves too to the right, and moving it by different amounts in each color channel, thus subtracting different amounts of light pollution. Generally, the red channel has the most light pollution, followed by green then blue.
Online I see many tutorials that say to align the histogram peak of the 3 color channels. That is WRONG. Aligning the histogram peaks suppresses the dominant color in an image. This is discussed in sections below: "Diagnosing Problems Using Histograms" and "White Balance and Histogram Equalization" where I show examples of the disastrous effects on image color. Steer clear of any site that tells you to do this to your images! What we really want to do is establish a good zero point (what is black in an image)?
Instead of aligning histogram peaks, the far better way to process night sky images is to align the start of the data (the lowest signal). This establishes a zero point. Of course the true zero point of the darkest parts in an image may not be color neutral, but because the signal is so tiny, it provides a best estimate of zero signal. In most night sky images, there are dark nebulae or gaps between nebulae to establish a best estimate of the zero point. Aligning histogram peaks not only does not establish a zero point (what is black), it can skew the zero point to strange colors. Indeed, online we see many images of the night sky with varying colors, commonly purple, and various shades of blue because many things in the night sky are dominantly red (like the Milky Way) and aligning histogram peaks shifts the image from red to blue as illustrated in the section below "White Balance and Histogram Equalization" and Figure 9 in that section.
For this image I used the following settings in Photoshop's Adobe Camera Raw (ACR) (settings in Figure 2): I used no dark frames subtraction because the Canon 6D has on sensor dark current suppression, so none are needed. No flat fields were used because the raw converter removes the light fall-off using the lens profile. The 5 panels, from left to right in Figure 2 show the settings I used:
Regarding the raw settings. The large negative Black level subtracts the overall light pollution level from the linear camera data, moving the remaining light pollution into the linear region of the standard tone curve that ACR applies. This enables further light pollution removal without strange artifacts. Key to this step is to not clip the low end, as some deep space light is in the tail of the histogram, especially around the dark nebulae.
Next, synchronize all images of the night sky in the raw converter and save them image as nine 16-bit tiff files.
Figure 3 is an example of one image, but I made 9 exposures. I used ImagesPlus to align and stack. Aligning the images to minimize drift between frames is important. If there is any shift when the images are combined, resolution will be lost. Stacking is the method used to combine the multiple frames into one image (e.g. average). In ImagesPlus, I selected image set operations, translate only and ImagesPlus produced a set of 9 tif files where the images are aligned extremely well. You could also do the alignment by hand in photoshop by stacking the images into layers and changing the opacity so you can see the image underneath, then moving one image to match up with the other. That is tedious and a dedicated astro image processing program is much faster and easier.
Then I combined the 9 images into one by selecting combine files in ImagesPlus. I used Sigma Clipped average and set the threshold at 2.45 standard deviations. Figure 3 shows the resulting image from combining the raw files. The mainly orange color is due to light pollution from sodium vapor street lights from the Denver metro area. There are many ways to combine images, for example, a simple average, median, high and low value exclusion average, and others. I find that sigma clipped average works best with all the astrophotos and different cameras I have used. Part 3e of this series compares some stacking methods.
In the image editor use the curves tool to further remove light pollution by moving the lower left point in the curves tool to the right for each color channel. (The left slider in the levels tool does the same thing.) Figure 4a shows an example using the image in Figure 3 (the output of the stacking) where only the green and red channels have light pollution subtracted from the entire image. Light pollution is minimum in the upper left corner and the curves work made the sky a neutral gray in the upper left corner. Note in the photoshop histogram tool, if a triangle in the histogram window shows in the upper right corner, the histogram is not current, and you must click on the little triangle to update the histogram after each adjustment in any editing step. Failure to update the histogram may result in clipping because one does not have accurate information.
The image resulting from the Figure 4a step is darker because we subtracted values from the image data. So the next step is to brighten it. Brightening will also magnify remaining light pollution and make the light pollution gradient from left to right more apparent. This is shown in Figure 4b. To increase brightness without saturating more stars, use the curves tool and make the RGB curve convex upward (Figure 4b). Where the slope increases, contrast is enhanced and small differences in intensity become more apparent (Figure 4b, burnt orange arrow). Similarly, where the slope decreases, intensities are compressed and contrast is reduced. The faint signals of nebulae are close to the sky level, so the shape of the curve in Figure 4b helps. Also helping is subtracting a small amount of red and green from the remaining airglow. The amount to subtract is found when the red, green, and blue histogram, lower left points light up. That neutralizes the darkest parts of the image, finding the best estimate of the zero level (this does assume some part of the image is black space).
The result of the curves tool stretch in Figure 4b shows the red, yellow, and blue nebulae surrounding the bright orange stars, Antares, starting to show,
Next select regions of airglow gradient, feather the selection and subtract as in Figure 4b. For example, these is a left to right gradient from light pollution from the Denver metro area, with maximum light pollution on the left. I made a selection aligned with the left edge and top to bottom as shown in Figure 4c. The light pollution gradient is embedded in the Milky Way brightness fall-off so removing light pollution here is complex. I estimated the gradient is slow from left to right so I feathered the selection by 999 in photoshop (Figure 4c). Then I subtracted using curves as illustrated in Figure 4c.
There is increasing airglow in the image from top to bottom (e.g. the bottom looks green in Figure 4c and 4d, and that is airglow). As above, I selected the airglow region, feathered it and subtracted it using curves.
The subtractions need not be perfect. It is better to not clip and under correct than clip and lose information. One can always correct more later. Curves work can be done in multiple iterations.
In subtracting light pollution, be careful not to subtract too much, or change color channel relative levels such that colors in deep sky objects are neutralized. For example, the full image histogram in Figure 4a - 4d still shows the red peak at maximum, green in the middle and blue lowest. The Milky way is reddish brown, so the histogram should NOT be neutralized, so making the histogram peaks line up should be avoided because the overall image should be reddish-brown. Aligning the histogram peaks or right edges has unintended consequences of reducing/warping other colors. For example, in other processing methods I have seen, neutralizing the histogram peaks in a case like this reduced the red channel, suppressing red nebulae (e.g. reducing hydrogen-alpha response--this is discussed below and shown in Figure 9).
I made a couple of additional iterations similar to the above strategy which resulted in the image in Figure 4e. The nebulae are starting to show quite nicely. At this point, the faintest parts of the image are quite low (note the histogram colors are aligned at the lower left in Figure 4e and close to the left edge. At this point do not do any more subtractions as there is a danger of clipping more data. Use the curves method shown in Part 3a for additional curves work.
Next stretch and enhance the deep sky stars and nebulae with an iterative process by following the method in Part 3a. If after stretching, if more light pollution and airglow need might subtracting, just do another curves run as described in Part 3a. Another step that works well in bringing out the nebulae in 4e is to do some saturation enhancement. For example, save the image in 4e and run saturation enhancement to see what improvements you can do.
As a final step, I sharpened the image with Richardson-Lucy image deconvolution using ImagesPlus with a 5x5 Gaussian profile, 8 iterations.
The image after all iterations of stretching and light pollution/airglow removal is shown in Figure 5a. It is an exercise for the user to try multiple iterations to get to the results from Figure 4d to 5a.
Larger images are:
1/3 resolution result (2 megabytes),
1/2 resolution result (3.3 megabytes).
One side effect to larger stretches of astrophotos needed to bring out faint detail is that lens flare is magnified making stars appear as small disks, both saturated and unsaturated stars. Astrophoto image processing software often includes star size reduction algorithms. I used the star size reduction tool in ImagesPlus to make stars smaller (Figure 5b). This has an interesting side effect: because stars are smaller, nebula stand out more. This allows one to change what you want to emphasize, stars or nebulae. I like the images in Figures 5a and 5b for different reasons. Another side effect of the star reduction algorithm seems to be a dark halo around stars in front of bright nebulae, and that I do not like.
Larger 1/2 resolution result (3.5 megabytes).
Well the internet is at it again. Many have tried my challenge below, but all too commonly they end up with a bluing of the stars from left to right. Some insist that the changing colors of stars to blue is real as one moves away from the galactic plane. To test the true color I used the Tycho 2 Star catalog of over 2.4 million stars down to past stellar magnitude 15 as described in Part 2b) The Color of Stars in this series. I split the region of the image in Figure 5 into 4 zones with zone 1 being on the left side of the image, working to the right and zone 4 is the right 1/4 of the image. I queried the Tycho data and plotted the histograms of the star colors in each zone in Figure 6. Not only do the star catalog data show that there is no bluing, notice zone 1 is highest in the blue and tends to be lower in the red. I also made maps of star colors as a function of position using the Tycho 2 star catalog data and the results are shown in Figure 7 of Part 2b) The Color of Stars. The star catalog data clearly show that, not only is there no bluing, there is a change to the red away from the galactic plane!
The conclusion is that if your processing is producing a bluing of the stars away from the galactic plane, it is an artifact of your processing. It is not real. This bluing seems to be at least partially responsible for reducing the red H-alpha signatures in people's processing attempts.
People are using their own methods on the raw image files. For example, here you can see results using other methods: Reddit Astrophotography and dpreview Astrophotography forum.
The original raw files are here on dropbox. Try and see what image you can make.
If you do the challenge you give permission for me to post a small version of your image here so I can discuss results.
First, it is fine to criticize my image; I do not claim it to be perfect or the best thing out there. I posted the image and this challenge to see if someone can show better ways to:
If you do the challenge you give permission for me to post a small version of your image here so I can discuss results.
People on the internet are pretty funny. One the one hand someone says to image other H-alpha targets that are more challenging and on the other hand they say the H-alpha in the challenge image is too faint. I have even been accused of boosting the red channel to invent the red H-alpha nebulae in the image. The traverse shown in Figure 7 shows that the H-alpha nebula is clearly present in the data before any enhancements were made to bring out the nebula. It is such a clear signal, it is surprising that some have trouble making it visible.
Then even funnier is how they point to other images made with CCDs, modified DSLRs, larger optics and hours of exposure in their efforts to criticize a simple 9-minutes of exposure with a little 100 mm f/2 lens. They are missing the whole point of the challenge. It is NOT to produce the best ever Rho Ophiuchus region image. it is to show the best processing results with whatever data one has.
I do not claim here that my processing of the challenge image is the best and only way. It is a new way, a method that I have been doing since about 2008. I don't claim it to be perfect, and I continually refine it. But the method does have advantages which are discussed in the other articles in this series. Of course, the specifics of how one implements the various tools of any method can have both positive and negative results regarding the final output.
Then it is even funnier that critics on the internet insist on many things, like blue star gradients away from the galactic center. The stellar photometry data show such a claim to be bogus. There are other claims, including that one needs bias frames and scaling dark current. But bias is a single value (2048 on the 14-bit camera raw data from the Canon 6D), and the camera has on-sensor dark frame suppression. Thus there is no need for bias frames, nor dark frames, regardless of processing method. Including these just adds noise and that reduces what faint signals can be extracted. That's a hint: do not use the dark frames in your processing if you try the challenge, and use a good modern raw converter. Figure 8 shows the difference in methods and using dark frames. ImagesPlus uses a traditional raw conversion to linear data similar to other astrophoto image processing systems. The image in Figure 8b has lower noise than the image in 8a if bad pixels are not included in the statistics. Dark frame subtraction (8a) removes bad pixels in traditional raw converters. Modern raw converters, like ACR (Figure 8c) use the bad pixel list in the raw file to skip those pixels during raw conversion. ACR also allows noise reduction at the raw conversion level (the amount of noise reduction is completely selectable by the user, so if you believe the noise reduction in Figure 8c is too much, simply choose a lower amount). Trying to do noise reduction on the images in Figure 8a or 8b has a higher chance of making a splotchy background due to the color noise (which I have seen in attempts by some people in this challenge).
It is the lower noise as illustrated in Figure 8c that enables one to pull out more faint detail. The noise reduction produces results similar to much longer exposure times with traditional methods.
Blue stars are popular. I really don't care if you want to manufacture far more blue stars that exist in nature. It does make for more colorful images and is certainly artistic license. But don't tell me I'm wrong if I try to process for a more natural color balance. Most post processing I have seen for this challenge so far has created a bluing color gradient that is a post processing artifact (or intentional), and people should know that these color gradients are impacting pulling out faint nebulae, especially the red H-alpha. Of course the "internet experts" don't agree. Fine--the photometric data does not support their claim that such blue gradients are real. Why this is important is that by creating the blue gradient, they are reducing red relative to blue and that suppresses red H-alpha.
Histogram equalization can suppress colors. The results in Figure 9 show different processing attempts. Which panels had histogram equalization, or some form applied? Some are really obvious, like f. But it is not just alignment. It is also shape. For example, two images, g, i have each channel with the same shape, but the green channel is shifted. That causes a color cast to magenta. Panel h is interesting because the red channel histogram encompasses the blue channel histogram and green at both the high and low ends. More can be discerned by selecting smaller portions of an image and examining the histograms for each channel. That can show regional histogram equalization, for example. Selecting just the background in an area can show what was done there, including histogram equalization. And this shows whether the image data are 16-bit tiff or 8-bit jpegs. The jpeg data might be a little noisier, but the trends and diagnostics are clear.
The dominant color of the Milky Way, from the star color to red/brown dust and hydrogen emission is red. Any histogram equalization in that situation reduces red relative to blue. Thus, the weak red signals, like that in the faint H-alpha nebula are pushed even weaker.
Astrophotographers can color their images any way they want. And changing color balance to show a broader range of colors in a scene certainly makes for a more interesting photo. But in applying algorithms, understand the consequences. Histogram equalization that makes a red background bluer reduces red. That means reducing the H-alpha response. The stellar photometry data show that the gradient away from the galactic plane has a greater number of red stars, not blue (see part 2b of this series: Color of Stars). The faint star background in Milky Way photos would show redder as one moves away from the plane of the galaxy for natural color. Histogram equalization in the many examples to this challenge I see online has created a blue gradient and many more blue stars (examples are seen in Figure 9, panels c, d, e, f, and h; g and i too but the magenta cast hides the blueing to some degree). It makes for a pretty picture, but is not natural and has the side effect of suppressing red H-alpha.
The red peak in the histogram panels a and b in Figure 9, about 2/3 to the right, is the light from the reddish Milky Way on the left side of the image. Panels c-i have suppressed the red so the peak in the histogram is not seen.
All images that I have seen in response to this challenge, including mine, could use more work. It is a tougher problem than it first seems, but is representative of issues astrophotographers face when trying to extract the faint info from an image regardless of the optics, cameras, and exposure times. My point of the challenge is to find weaknesses in any workflow and improve on them, including mine.
Following on the above discussion of histogram shape, it has become clear that many astrophotographers at this time (2015) are performing auto white balance of some form, including histogram equalization. This is destroying natural color and often results in suppression of red hydrogen-alpha emission.
To show the effect white balance methods have on images, I show the effects of different processing in Figures 10a, 10b, and 10c. Clearly the auto color and histogram equalization (Figures 10b, and 10c) have destroyed the beautiful colors of the sunset. This will happen with ANY image that has a dominant color--the dominant color is suppressed in order to enhance other colors. Note that in Figures 10b, and 10c, we see more blue. The auto white and histogram equalization suppresses the dominant color and boosts weak colors. The histogram equalization in Figures 10c has also enhanced green, making green clouds! Clearly these colors are not real, nor did I see any such colors at the scene. Histogram equalization and auto white balance should be avoided in nightscape and astrophotography, and in my opinion, all photography where you want natural colors.
See Figures 9a and 9b in Aurora Photography for the effects of auto white balance on aurora images.
Astrophotography image processing using modern raw converters and simple image editors, primarily using the curves tool, can extract a lot of information from images containing significant light pollution. Specialized software is still needed to align and combine the multiple exposures.
A number of people have taken the challenge and posted their results. It has become clear that commonly applied processing methods used in astrophotography suppress red and H-alpha in this image. The problem seems to be application of a histogram equalization step in the processing work flow. This is like auto white balance and if you applied such a step to a red sunset image, the result would be a boring sunset with little red.
The implications shown here are that image processing methods have a lot to do with H-alpha response and in extracting faint signals from the data. The Milky Way is actually dominantly yellowish to reddish, so the auto white balance reduces that dominant red. Maybe the use of modified cameras is not needed to the extent that people think!
References and Further Reading
Clarkvision.com Astrophoto Gallery.
Clarkvision.com Nightscapes Gallery.
The open source community is pretty active in the lens profile area. See:
Lensfun lens profiles: http://lensfun.sourceforge.net/ All users can supply data.
Adobe released a lens profile creator: http://www.adobe.com/support/downloads/detail.jsp?ftpID=5490
More discussions about lens profiles: http://photo.stackexchange.com/questions/2229/is-the-format-for-the-distortion-and-chromatic-aberration-correction-of-%C2%B54-3-len
Compare to other Images of this scene
The Dark River to Antares Astronomy Picture of the Day by Jason Jennings. http://apod.nasa.gov/apod/ap150222.html
H-alpha modified Canon 6D + 70 - 200 mm f/2.8, 36 minutes of exposure, ISO 1600, by Tony or Daphne Hallas: http://www.astrophoto.com/RhoOphiuchus.htm
Dedicated CCD: Exposures: H-alpha: 5X24 minutes, RGB: 8X5 minutes each, 200 mm focal lens at f/4, 160 minutes total exposure, by Michael A. Stecker: http://mstecker.com/pages/astrho200mm-HRGB5smJRF.htm. This image is 21 times the exposure from the subject compared to the 9-minute image above.
Canon 60da (that is an H-alpha astro modified camera), 70 mm focal length lens at f/4, 51 minutes exposure, ISO 1600 by Jonathan Talbot: http://www.starscapeimaging.com/page88/index.html. Exposure relative to the 9-minute image above: 0.7x.
The Night Photography Series:
| Home | Galleries | Articles | Reviews | Best Gear | Science | New | About | Contact |
http://www.clarkvision.com/articles/astrophotography.image.processing
First Published July 6, 2015
Last updated Fenruary 28, 2016