Astrophotography Part 6 of 6: Reducing Image Noise
Tips for Reducing Noise
By K David in Articles and Tips on Aug 11, 2017
High-quality digital cameras with good high-ISO performance combined with fast, high-quality optics from makers like Pentax, Samyang, Sigma, and Tamron are taking astrophotography out of the enclaves of astronomers and putting it in the hands of hobbyists, too. This six-part series examines various elements of astrophotography.
This article looks at ways to reduce noise in digital images. Using best capture practices, software, and editing techniques, a significant amount of noise can be removed from overnight photos.
To revisit the rest of the articles in this series, here are links and their subjects:
- Series Announcement
- Article 1: Planning the Shoot
- Article 2: Equipment Selection
- Article 3: Making a Barn Door Tracker
- Article 4: Using Film
- Article 5: Working in Post
Digital noise occurs in every image. if you haven't heard this term before, it will appear often in this article: signal to noise ratio. Signal is the part of the image that we want to keep and noise is the part of the image that degrades the signal. All digital images have noise. All film images have noise, too, but in film it's called base fog or dichroic fog, depending on how the fog occurs.
To understand how noise degrades images, it helps to use an analogy. A glass of water is a common analogy and works well.
Exhibit 1: Like Water for Signal
Imagine that the water (blue) represents the signal. A short exposure receives less signal. So a five-second exposure might show you a bunch of stars, but it also gets only a relatively little amount of signal. Relatively speaking, a 30-second exposure will receive a good deal of signal compared to the five-second exposure.
Exhibit 2: Make Some Noise
All images have noise. Your camera makes noise because it runs on an electrical current that causes some noise to occur in the signal. Your images have noise because you're using the sensor at a higher ISO than the native sensor ISO (200 for Pentax cameras.) Your images have noise because stray high-energy particles smack into your sensor and sometimes cause a noisy pixel. Your images have noise because the sensors get hot and the pixels record data that isn't there. Your images have noise because image media are a human invention and they are, by definition, therefore imperfect. In film, noise occurs when silver crystals spontaneously expose themselves. It also occurs when chemistry causes film grains to develop when they weren't exposed, or if the chemistry allows silver to re-deposit on the film's surface during fixing. So we'll have some noise and the amount of it is somewhat consistent between images (though it increases with prolonged sensor use, especially overnight.)
Exhibit 3: Images with Noise
With a short exposure duration, noise has a noticeable and unpleasant effect on images. With longer-duration photos, it has less of an effect. Two methods exist to reduce noise in an image: reduce the noise or increase the signal. This article will examine ways of achieving both aims. The better of these two options is to increase the signal. But noise reduction also has a place in astrophotography. Used in concert, noise reduction and signal increases can result in images with very true, dramatic colors and details. And yes, all of the techniques we discuss here work in different ways can be combined.
Types of Noise
To understand noise reduction techniques, we need to also understand types of noise. Not all noise is created equal and not all can, or should, be handled the same way. Digital noise exists in two types: chromatic and luminance.
As their names imply, the two types of noise affect different things. Chromatic noise affects colors. This type of noise manifests as mis-colored pixels, typically very mis-colored, in a sea of properly-colored pixels. Luminance noise is noise that affects a pixels shade, but not the color. Luminance noise typically increases saturation but could manifest as a darker hue.
Clicking the image above will open it in an overlay window for better detail viewing. This is a 100% crop (at a 1080 monitor resolution) from a Pentax K-3 of the Whirlpool Galaxy. This was taken through a Takumar 500mm f/4.5 at f/5.6 at ISO 200. This crop shows chromatic and luminance noise.
Exhibit 5: Sensor Barf
Luminance noise, exhibited in the image above, looks like sensor barf. And basically, it is. Instead of a flat, dark sky the night looks like the mottled coloration of a frat house carpet on Sunday morning. This is severe luminance noise.
Exhibit 6: Chromatic Noise
In this image, I circled some of the hundreds of spots of chromatic noise in the image. The chromatic noise are hot pixels that look like, but aren't, stars. These occur when pixels tell the camera that they received 100% (a full 255 on the RGB scale) of their designated color.
Sensor Sensitivity (ISO) Settings Do Not Increase Noise
From a technical perspective, the sensitivity you set your sensor to does not create noise. Digital noise from higher sensitivities is due to image gain, the amount an image is lightened due to increased sensitivity. Increasing sensitivity appears to add noise to images when using an automatic exposure mode because the shutter speed is duly decreased, tilting the signal:noise ratio away from the signal side and to the noise side. In astrophotography, because we use manual exposure modes, we can shoot at significantly higher sensitivities and have less noise than we see with lower sensitivities.
It sounds paradoxical, but you'll see shortly a demonstration with the Pentax K-3 that shows no meaningful noise difference between an astronomical photograph taken at 400 ISO and another taken at 12,000 ISO. This is because, since at least the K-5, Pentax has used ISO invariable sensors. This section contains an explanation of that later on.
The primary mode of noise introduction in astrophotography has to do with shutter speed, specifically related to exposure time and whether or not an exposure is suitable. The secondary mode relates to aperture. Aperture can cause noise by limiting light to too great an extent: underexposing the image. Also, wide apertures can allow sideways glare through the lens to the image media. Also, using too long of an exposure will introduce noise through sensor use and processor limitations. Key to reducing noise is to understand how your sensor works.
So to reduce noise in capture, use a longer shutter with a wider (but not fully open) aperture. That will allow you to use higher sensitivities without increasing noise (in fact, noise will likely be reduced.) But work to avoid drastic overexposure as highlight recovery is much more difficult for Pentax cameras than is shadow recovery.
Exhibit 7: 3200 ISO, f/2.8, Two Seconds | ISO Invariance Point Key Frame
Some camera makes show greater image noise at lower sensitivities. Pentaxes, we read on the Internet, are susceptible to this. We decided to test common Internet wisdom, however, and our results surprised us. The test we performed on a K-3 demonstrated the converse of Internet wisdom: Pentax cameras are not susceptible to noise gain due to ISO. In fact, Pentax DSLRs, at least from the K-5, can be used in a technique called ISO-free photography. This is not a term we agree with, nor do we condone its use. But if you want to research this further, ISO-free photography is a useful Google search term.
We know that Pentax cameras, at least from the last five years, are what's called ISO invariable. This means, in theory, someone could shoot long-exposure images on a Pentax at most any ISO with no recognizable change in noise following adjustment in post. If two Pentax photographers were next to each other photographing the Milky Way and one used a Pentax at ISO 3200 and obtains a well exposed image, the other Pentax photographer taking the same shot at ISO 800 and pushing the raw file two stops in post could obtain the same image with no observable difference in noise. ISO invariance means that Pentax cameras will be forgiving of exposure errors when raw files are processed.
From a technical perspective, this means that more of the final image is derived from the camera's computer than from the sensor. ISO invariance is not a measure of sensor or camera quality, but having an ISO invariable camera means the gear is well suited for night photography.
About 99.9% of the time, this information is only important in low-light work. In practical terms, it means that a Pentax user with a good raw editor can commit more exposure sins with overnight photography than someone with an ISO variable camera. In short terms: Pentax cameras are very forgiving of low-light underexposure.
Canon cameras, for instance, are ISO variable cameras that exhibit noise gain when shooting overnight images at low ISOs. (Note that the amount of ISO variability can differ between different models by the same maker and Canon is not the only maker with ISO variable sensors.) Also, ISO variability tends to decrease or stop at some point when you raise the ISO high enough. It's worth experimenting with your DSLR to find out the cut-off ISO. You can do this by running a series of test images (raw is mandatory for this) at a consistent aperture and shutter speed adjusting only the ISO. If you have a Pentax K-3, we've done that for you and the results are below.
Some things, first, if you want to perform this test yourself. Turn off your in-camera noise reduction and set a specific white balance so that the white balance does not change between images (we neglected to set the white balance in our test and the resulting cloud color cast changed between shots.) Shade or daylight are good white balance options. Artificial lighting white balances may degrade results.
In the above image, which can be clicked on for a larger version, we captured nine frames from ISO 200 (upper left) to ISO 52,000 (lower right) in full-stop increments. The meter was set to provide an accurate exposure at ISO 3200 with the 40mm pancake lens at f/2.8 and a two-second shutter speed.
To do this yourself, use ISO 1600 or 3200 as your key frame ISO. If you have an older DSLR with a lower sensitivity range, use the middle ISO setting. You'll want to manually expose to obtain an ideal exposure at your key frame ISO. So if you're outside and a proper exposure at your key frame ISO is five seconds, set your aperture and shutter speed manually and then shoot a series that stretches from at your lowest ISO to your key ISO and an equal number of stops past your key ISO. We used 3200 as our key frame because it allowed four stops higher (up to 52K) and four stops lower (down to 200 ISO) and didn't stretch past the sensor's native ISO. This also gave us nine images to work with and that made the two panel exhibits much more aesthetically appealing.
In post, you'll manually adjust the raw files (exposure compensation in Photoshop CS6's raw editor, other photo editors call the tool different names) to balance each exposure. To do that, let's assume that you used ISO 1600 as your key ISO. In post, increase the exposure on the 800 ISO shot by one stop. The 400 ISO by two, and so on. For the 3200 ISO shot, decrease exposure by one stop, the 6400 ISO shot by two, and so forth.
With an ISO variable camera, low ISOs will exhibit significant noise, reducing as the ISOs increase. At some point, with an ISO variable camera, you will see no discernible increase is noise between the samples. That point is an ISO variable camera's ISO invariance point and the lowest ISO setting with invariance ideal ISO for astrophotography. In other words, pushing your camera in-body beyond that ISO point will provide no discernible benefit in light capture.
For an example of ISO invariable cameras, we look at Pentax, Fuji, high-end Nikon, and Sony cameras. These bodies show no post processing noise gain using this test even at ISO 100. Exhibit 9 presented two similar images. The first was taken at ISO 12,000 and the second at ISO 400. They look almost identical in terms of noise following raw exposure balancing to match their exposures to the ISO 3200 key frame. If you'd like to check all nine frames, click on any of the images below and the pop-up box will allow you to click through the gallery. The test shows that the Pentax K-3 is ISO invariable. This means that a great deal of raw editor noise correction can occur and that higher and lower ISOs can be used to great advantage, including to force over- and underexpose shots out of the camera for correction in post.
Pentax Auto Noise Reduction
Pentax cameras come with automatic noise reduction. This option increases processing time between photos, has some ramifications for sharpness, but can also reduce image noise. The in-camera noise reduction feature's efficacy varies between models with newer models with current firmware typically having the best in-camera noise reduction. However, like all functions that can be performed in-camera, the results tend to be less appealing than when the same process is performed outside of the camera in post with a high-quality software. Specifically, external software tends to handle image artifacts better than in-camera noise reduction.
In-camera noise reduction functions by taking a second image, a dark frame, immediately following the photo that was just taken. This allows the camera to find the noise-causing pixels and remove or alter them in the original image. This means, yes, that the exposure's time is doubled, which ruins any attempt at using in-camera noise reduction for star trails.
Dark Frame Technique
However, there is a technique similar to in-camera noise reduction that can be performed in post with changing raw file data or doubling exposure capture time. The dark frame technique allows you to take one or two photos at the start and end of a sequence or shoot to use for noise reduction. As an added bonus, this technique can be used for any photography, not just astrophotography. Dark frames can reduce noise, sensor amp glow, color bias, and a number of other issues related to the physics of how digital imagery works.
To take a dark frame, take a photo with the lens cap on at the same settings (ISO, exposure time, and ambient temperature) at the image or image series. For overnight work, taking a photo at the start and end captures differences in ambient temperature and sensor. More on this shortly. First, let's look at three images.
Exhibit 11: Dark Frame
This is a standard dark frame. It doesn't look like much because my camera had been resting for two days before I took this. So let's see what it looks like with a number of levels adjusted (in post) to illustrate what's hiding in the darkness.
Exhibit 12: Dark Frame Exaggerated
So at what point does this start showing? Here is a 118-second exposure with the camera at ISO 3200.
Exhibit 13: ISO 3200 Dark Frame
At ISO 3200, the sensor amp noise on my K-3 with a rested sensor becomes very apparent. To further illustrate the differences between the above images, here are 100% center-frame crops at ISO 200 and ISO 3200.
Click either image to see a pop-out window with a larger version and switch between them to see the noise difference. So now that we know what a dark frame is and how to obtain one, let's examine how they can be used to reduce image noise.
StarStax, ImageStacker, and StarTrails all have the capacity out of the box to process dark frames. In Photoshop, the dark frame is one of the final steps in image generation. Place the dark frame(s) as a layer (or layers) over the image you want to correct and set the layer blending to either subtract or difference, whichever looks better to you. Then adjust the opacity level until the result looks good.
In Photoshop, if you collect a dark frame before and after the shoot or image series, you can use both of the dark frames to further reduce noise. Using two dark frames will help reduce the random noise generated in images in addition to hot pixels and amp glow. This is because even dark frames are slightly different; they capture different noise and artifact data. So by combining them it's possible to achieve greater noise reduction. In addition, to further refine the results, using different blending modes with each layer and different transparency levels can help further reduce noise.
If you forget to take dark frames during a shoot, you can use stock dark frames taken with your camera at some point either days, weeks, or months before the shoot. This is less successful, however, than using dark frames taken before and after the shoot. Frames taken at the time of the shoot reflect sensor conditions when the photos were taken. Age, temperature, and immediate use or rest can affect sensor performance and noise.
Image stacking, as a process, is a subject we've tackled often in Pentax Forums articles, including in the previous installment of this series. Stacking multiple exposures reduces noise because noise is random so, over a series of images, the likelihood of the noise manifesting in the same pixel throughout the series is infinitesimally small. So blending the images together in a stack means that the average data for each pixels eliminates the noise on individual images. For more information on stacking, check out this article, which provides an in-depth resource into blending modes and their benefits and details.
There are a number of third-party software programs available for noise reduction. Some work better than others. Here at Pentax Forums, we recommend Topaz DeNoise. Here are two very different photos that we processed with DeNoise. The first shows Messier Objects M81 and M82 taken with a Pentax K-3 and Takumar 500mm f/4.5, We'll show the originals and then the DeNoise versions in both the full frame and a close-up crop.
Exhibit 16: M81 and M82 Original Image
Exhibit 17: M81 and M82 Image with Topaz Labs DeNoise Processing
The above images show that third-party software has a lot of potential for noise reduction for both luminance and chroma noise. The result is that images exhibit truer and more consistent colors with insignificant detail loss.
The above images show the benefits of third-party noise reduction software in a close-up crop. Click on these photos to examine and switch between larger views.
The following four photos present the same process shown above with a sun photo. This shows that even in an inherently low-noise image (contrasted with the very-high-noise image above) third-party software provides a number of benefits and advantages.
Exhibit 20: Solar Photo Original
Exhibit 21: Solar Photo Following Topaz Labs DeNoise Processing
On a large scale, noise reduction on a low-noise images helps to true the color s and reduce things like the unwanted glow around the sun. Here are up-close crops of the above images. As above, click on these to see larger versions.
If you would like to purchase Topaz Labs DeNoise and benefit Pentax Forums at the same time, follow this referral link: Topaz Labs DeNoise.
There are other options, too, for third-party noise reduction software. For a detailed comparison of seven specialty noise-reduction software and two Photoshop blending modes, check out this article by Fstoppers.
Noise reduction begins before the shoot, continues through the shoot, and ends in post. To help control image noise, here are some tips and best practices. Not all apply all the time, but each of these helps improve results in different ways.
- Shoot in winter if possible; the cold reduces digital sensor noise
- Use a sensor that's rested and has cooled off prior to the shoot
- Shoot at a your camera's ISO invariance point
- Low ISO can provide greater noise than high ISOs for astrophotography
- Pause between shots unless you're shooting a star trails image
- Use vivid, landscape, or reversal for your JPEG output as they raise contrast (hiding some noise)
- Shoot raw, regardless of JPEG setting
- Shoot with all the noise reduction corrections in the camera off as they are less powerful than those used in post software; the exception being dark frame cancellation which is a good in-camera tool for some situations
- Manually set your color temperature; these settings are a good baseline (but your experimentation will help refine them for your camera) and can be adjusted in your raw editor if not your camera
- Color temperature in the low 4,000-range
- Magenta of -1 to -4
- Green of -1 to -4
- This helps hide noise without impairing later color correction
- Use an inexpensive fast lens like the Samyang 10mm 2.8 or 24mm 1.4
- Calibrate infinity ahead of time
- Manual focus is ideal for astrophotography anyway
- Get your in-camera settings right and make sure that they're consistent across shoots, this lets you know what to expect in post and develop a routine for your work
Thank you for following along throughout this article series. We recognize that we took far longer in getting it to you than originally planned. We hope that the content has met or exceeded your expectations and that what you've learned here will improve your photography and engage your interest in photographing the sky. We've appreciated your comments and feedback on this series and are eager for your feedback on what we can do to improve future articles here on Pentax Forums.