Why Is Noise Reduction Such a Big Deal?
Noise is a big consideration in shooting pictures at night. We push our cameras to capture the best details possible in dark conditions and the result can often be a little noisy and that can be especially noticeable when we go to print. What can you do about it? Well, part of the answer lies in the shooting techniques but there's more on the post-processing side as well. In the end, photographers have some good options available to them.
One of the challenges in using noise reduction is that there are so many options available to us. In-fact, with all the different approaches available for reducing noise in nightscape images, it can be a downright confusing to know what to use when. So Darren and Mike got together and tested some of the most common methods available to determine what would be best to recommend and when.
We're not going to pretend that everyone or every image needs noise reduction though. In-fact, some people say that the details of a night image come in-part from the noise and if you remove all of the noise, you'll lose a lot of the details. We generally find that the noise is an issue mostly when we're printing an image and that's when we want the absolute best image quality possible. So in the end, we recommend using noise reduction only in select cases where it's needed and even then, we like doing it with a light touch.
Types Of Noise
In night photography, we see a few different types of noise and for this discussion, we're going to focus on two of them (Hot Pixel Noise and Random Noise). And since some noise reduction techniques often work well on one type of noise but not on another, it's important to understand the differences between them. Chromatic Aberrations and Banding Noise are two more types of noise you may also see but we won't be focusing on them in this discussion.
- Hot Pixel Noise (aka Fixed Pattern Noise)
This can be seen in a variety of conditions and is most noticeable in warmer ambient temperatures and when the camera sensor has been used for some time. Of the two types of noise, this is the easier one to remove. This type of noise appears as spots with bright colors and in a sequence of images, they tend to appear in the same location on each image.
- Random Noise
This can also be seen with different settings but it's most noticeable when you use the higher ISO settings and it's especially bad when an image is underexposed. It appears as splotchy pixelated areas in the dark voids (between the stars) that would otherwise be smooth. It tends to be harder to remove than the hot pixels. The position of this noise is random so the splotchy pixels won't appear in the same position in all of the images captured in a sequence.
Adding Data vs Single Exposure Noise Reduction
Normally when people talk about noise reduction, they refer to approaches that can be used on the post-processing side. There are some notable options on the shooting side as well so we're going to take a look at those too.
- Adding Data with Multiple Exposures
This is where we add another image that has more pixel information. Whether it's just another exposure with the same settings (aka light frame) or another approach, astronomers have been using some interesting techniques for decades and some of them can help us as well. What separates the approaches in this area vs the techniques used in post-processing on a single exposure is the fact that adding data tends to reduce noise without also giving up some details. So extra attention is deserved in this area since using a good shooting technique usually translates to the best image quality and this route is often preferred over trying to extract a high quality result from noisy image.
- Noise Reduction in a Single Exposure
We may have fewer options to choose from but they're still pretty good thanks to the power of Photoshop and some third party tools from Nik and Topaz. Some of these options are really impressive especially for the hot pixels!
- Priorities: Reduce noise without sacrificing details
- Scores: 1-10 with 10 being excellent
- Image Samples: Shown at 100% with some brightening to make the noise easier to see
- Equipment: Images used in testing were captured a Nikon D810, D800, D700, and D300. Your results may vary depending on the manufacturer and model of the camera you're using and the content of the scene you're capturing.
Techniques For Adding Data
- Smart Object With Median Stacking
This is where we take a sequence of images and blend them using "median stacking mode" which essentially throws out the bad values and only keeps the good ones. The result is a visible reduction in noise although it takes some work to get there.
The stacking process part of it, but don't forget that for the final image, we often like to blend in a long-exposure low ISO foreground. That's where the reward comes at a price with time and skills on the post processing side.
Median Stacking deserves a little more mention so we wanted to take it a little further. To do it, you'll need a sequence of exposures, Photoshop (cs6 or newer) software and skills, and some time. The benefit is reduced noise in the stars and this is the only approach that doesn't sacrifice details. And while the foreground is one way to see what it does for noise, we also wanted to show what it does for the sky. Keep in-mind that the more images we use, the better the noise reduction. This image only uses 7 frames.
- Lighten Blending Mode
Simply changing the blending mode for a sequence of images has some success in reducing noise and it's easy. In just a few moments, we can turn a sequence of noisy images into a single one with less noise. We noticed though that while this technique works fairly well with random noise, it works very poorly with hot pixel noise.
- In-Camera Long-Exposure Noise Reduction
During long-exposures, a lot of short exposures, or when the ambient temperatures are very warm, we tend to have more of the Hot Pixel noise. When that happens, we may find some benefit from using our cameras' in-camera long-exposure noise reduction. When we do this, we'll find that the camera stays busy for twice the amount of time as what we set for an exposure time. This happens because the camera is actually shooting a second exposure without letting any light in so it can measure the amount of heat coming from the sensor. Then it removes the noise coming from this heat from the first image.
While we did plenty of testing with in-camera long-exposure noise reduction, we found it's difficult to compare what it looks like with and without having it enabled because of the varying temperatures of the camera's sensor. Besides that, running multiple long exposures with a single camera is a bit of a challenge. So while it's difficult to test it in a side-by-side comparison, the scores were consistent. It did well with hot pixel noise (score: 8) but not so much with the random noise (score: 2).
- Dark Frame Subtraction
Using some astrophotography terminology, the primary exposures we typically capture are called light frames. After that initial frame, you can measure the amount of heat coming from your sensor by shooting a dark frame - most often captured with the same settings as your initial exposure only with the lens cap on to block the light. This is the manual approach to the in-camera long-exposure noise reduction mentioned just above. It's common with telescopes and has some benefits with wide-field astrophotography as well.
In testing, I found I didn't really care for using Dark Frame Subtraction. It had some unintended results with the image so I found most people who use dark frames (mostly astronomers) have more than one. Using a few of them with reduced opacity gives some improved results, we've heard. We don't have time for that in the field (shooting multiple long-exposure frames with the lens cap on) so we don't recommend it.
Techniques For Noise Reduction In A Single Exposure
- Median Noise Reduction
1. In Photoshop, copy the image layer
2. With the copied image layer selected, click on Filter, Noise Reduction, Median, 1, and then OK
This reduces both hot pixels and (to a lesser degree) random noise in the rest of your image. In many cases, this may be preferred (to address both types of noise). If your preference however is to just address Hot Pixel Noise with no chance of losing details in the rest of your image, the next option (Median Noise Reduction With Pin Light Blending Mode) bay be the better choice..
- Median Noise Reduction With Pin Light Blending Mode
1. In Photoshop, copy the image layer
2. With the copied image layer selected, click on Filter, Noise Reduction, Median, 3 (or 4), OK, and then set the blending mode to Pin Light
This technique does a great job of reducing the hot pixel noise without any effect on random noise or the rest of the image for that matter. It's very effective on its own or in combination with another approach for reducing Random Noise. But for the purpose of reducing Hot Pixel Noise without sacrificing any details, this approach looks terrific.
- Dust & Scratches Noise Reduction
1. In Photoshop, copy the image layer
2. With the copied image layer selected, click on Filter, Noise Reduction, Dust & Scratches, and then OK
This reduces hot pixels with minimal effects on random noise or details in the rest of your image. It works very well but it usually takes second place behind Median Noise Reduction with Pin Light Blending Mode.
- Nik Dfine
Features third party noise reduction tool called Nik Dfine primarily to reduce random noise. There are a lot of options and variations in how much noise reduction is applied. This technique can lead to a loss of details but the controls allow you to do a pretty good job of reaching a balance between soft & sharp.
- Topaz DeNoise
Of the third party noise reduction tools, Topaz DeNoise is arguably one of the better ones. Just like Nik Dfine, there are a lot of options and variations in how much noise reduction is applied. At times, I found DeNoise to be a bit tougher to control though as it often applied more noise reduction than I wanted - leading to a loss of some details. As a result, I found Nik Dfine a bit easier to apply a soft touch.
TEST Scores & CONCLUSIONS
|Adding Data With Multiple Exposures||Hot Pixel Noise Reduction Score||Random Noise Reduction Score|
|Smart Object with Median Stacking Mode||3||9|
|Stack with Lighten Blending Mode||1||6|
|In-Camera Long-Exposure Noise Reduction||8||2|
|Dark Frame Subtraction||5||2|
|Noise Reduction In A Single Exposure||Hot Pixel Noise Reduction Score||Random Noise Reduction Score|
|Photoshop Median Noise Reduction||8||4|
|Photoshop Median Noise Reduction with Pin Light Blending Mode||9||2|
|Photoshop Dust & Scratches||8||2|
|Nik Dfine 2||2||7|
|Topaz DeNoise 5||2||6|
- Testing confirmed that the best way to reduce Hot Pixel Noise is in post-processing - specifically using Median Noise Reduction with Pin Light Blending Mode. The results here are so good that we prefer it and recommend it over in-camera long-exposure noise reduction. The in-camera long-exposure noise reduction is usually effective in removing hot pixel noise but we can get similar results using the post-processing tools and we don't have to give up use of our camera while the noise reduction processing takes place.
- Testing also confirmed that the best way to reduce Random Noise is to add more data through multiple exposures. It's simply the most effective way to reduce random noise without sacrificing the details and the results are very impressive in both short or long exposures. If you only shot a single exposure, Topaz DeNoise is the best alternative knowing that it will only give marginal improvement.
- This isn't really new, but one more conclusion we get out of the testing is that it confirms how much easier it is to remove hot pixel noise than random noise. With this in-mind, we stand by the approach of using a long-exposure low-iso shot for the foreground. We use it on its own as an image that shows the stars trailing or we blend it with one or more shorter exposures for the sky. The lower ISO shots with a longer exposure time may have some hot pixels but they're easy to remove. The benefit of this approach is that it usually results in a foreground that's low in random noise which translates to good image quality in the foreground.
Recommended Techniques For Noise Reduction
- Hot Pixels Noise
- Median Noise Reduction with Pin Light Blending Mode
- Use this either for single exposure on its own or in a long-exposure low ISO image (planned for blending one or more images of the sky)
- Random Noise
- Stack Multiple Exposures With Median Stacking Mode when you can
- As an alternative, use either Nik Dfine or Topaz DeNoise
- Plan On Some Variation
- Keep in-mind that with different content in each image, results will vary.
- We recommend trying the top techniques for the type(s) of noise you're trying to reduce, reviewing the results at 100%, and then pick the one technique (or combination of techniques) that looks the best.
If you have comments or suggestions on more techniques, please let us know! We're very interested to hear your thoughts.