Noise is the sleet storm Satan drenches our photos in when we stupidly leave our tripod in the trunk thinking VR will save our lazy butt, but instead we end up shooting at quadruple digit ISOs. In this article, we will take a look at a couple of techniques on how to reduce noise and how to avoid it in the first place.
That grainy look to our photos is the punishment for our lack of photographic purity and wholesomeness. So how can we go about trying to minimize noise in our images and fool people into thinking we are without sin? I’m glad you asked.
My girlfriend has this nasty habit of taking better photos than me. To fix this problem, when we see a great subject like this Burrowing Owl. I hand her my D7200 set at White Pelican in Flight settings (-1 stop exposure compensation with Auto ISO enabled). Before she knows it, she’s already at 10000 ISO and still underexposing as seen above. Ha, I win! Talk about a recipe for noise stew. Fortunately, she’s kindly let me use the resulting file to demonstrate some noise reduction tricks and I only have to wash all the dishes for the next ten years.
Here’s a version I tweaked in Lightroom. It looks minimally noisy to me and get this, I didn’t nudge the sliders in the noise reduction panel at all to get this result.
Before we get into how to minimize the appearance of noise in our images let me state here that noise is one of the least important components when it comes to an image’s strength and appeal. Good composition, alluring lighting, proper exposure, appropriate focus, astute timing, etc. etc. are all much more important to a photo’s success. Noise is the last nitpicky thing I worry about when working on a photo. Some noise can even be desirable, for instance if you’re trying to achieve the gritty B&W photojounalistic look that came from the days of Tri-X film (very popular in the 60s and 70s). That said, at a certain point, noise can get so bad it degrades image quality by obliterating fine detail.
This article is written assuming we are capturing RAW images, not jpegs. In it, we’ll discuss controlling noise both in camera and when post-processing in Lightroom. Further noise reduction in other software programs such as Noise Ninja or Topaz DeNoise is beyond the scope of this tutorial.
1) Noise Basics
Noise comes in two flavors; luminance noise and color noise.
Luminance noise gives the grainy look above.
Color noise gives multicolored speckles.
Noise is most often the result of amplifying the signal (in this case the data collected by each photosite on your camera’s sensor) to output a file of the desired apparent brightness. As light levels drop and each photosite gets less information in the form of photons landing on it, to achieve a file that doesn’t look underexposed, we need to amplify the signal by boosting the ISO. The higher an ISO you shoot at (assuming you’re shooting the same camera), the greater the noise in the resulting files. Sensors produce the cleanest results when shot at base ISO (the lowest numbered ISO on the camera).
Smaller individual photosites (common on high megapixel crop-sensor cameras, many mirrorless offerings and point-and-shoots) get hit by fewer photons per exposure than bigger photosites (found on lower megapixel full frame DSLRs and medium format digital bodies) hence have less information to send out and thereby need to amplify that signal more to achieve the same level of brightness. This is why the 16mp full frame D4s with its Rubenesque photosites is much less noisy at the same ISO than the 36mp full frame D810 with its anorexic photosites.
Here an example from the creepy Iron Door Cave in the Mojave Desert. Shot at 6400 ISO with a full-frame D4s, it suffers from very little noise – even the dark areas look fairly clean and this with some shadow recovery applied. And here is the original file with no shadow recovery applied and at 100%
Contrast this with a crop sensor D7200 image at 7200 ISO at 100%. The D7200 has much smaller individual photosites.
This file is untweaked. The background is noisy. However, we should realize when we zoom in to 100% on a higher megapixel file than a lower one, we zoom in tighter and this actually increases the appearance of noise. If we shot the same scene on the two different cameras and zoomed to the same final composition the difference would not be as great, but the D4s would still win.
Noise can come from amplifying the signal in camera by boosting ISO, or later by boosting the signal during post-processing, for instance to recover shadow detail.
2) Avoiding Noise in Camera
The best way to avoid noise is to shoot at the lowest ISO you can get away with. This might mean using a tripod or perhaps supplementary lighting (flashes, etc.). The more photons that hit your sensor (from a longer exposure), the more info each photosite collects and less amplification required to output the desired signal (how bright or dark that part of the photo will look).
In camera high-ISO noise reduction is only applied to jpeg outputs. If you are saving the images from your camera as RAW files, turning high-ISO noise reduction on will have no effect – you will have to do your own noise reduction in post.
- Avoid underexposure. Increasing exposure in post requires amplifying the signal and is an express ticket to Noiseville.
- Exposing to the right (ETTR) is a trick to get the maximum amount of info in a file so one needs as little signal boosting in post as possible. ETTR involves intentional overexposure of the initial file (just short of blowing out the highlights), then dialing back the brightness in post. A big ETTR benefit is less apparent noise in shadows. When I use ETTR tactics, I simply bracket shots until I find the brightest exposure that gives me a file without highlight blinkies. For a more detailed exploration of ETTR, click on Spencer Cox’s article here.
- Compose tightly in camera. Cropping in post will magnify the appearance of noise. Better to step a bit closer to your subject or zoom in if you can. Shoot multiple versions in camera so you can take the one that needs the least cropping in post.
- Favor light backgrounds over dark backgrounds. Noise is most apparent in the darker parts of a photo.
Note how noisy the midtones and shadows in the background are. Note how little noise is apparent in the brighter branch.
This background tip is one I put low priority on as choosing an appealing background is more important to me than choosing a background just to minimize noise.
Realize that highly detailed parts of a photo can do a great job at hiding noise. Such areas might be fur, hair, feathers, grass, sand, etc. Conversely, smooth areas, like skin or sky, will make noise more apparent. Looking at the bluebird above little noise is apparent on it’s finely textured back feathers, however the broader smoother tail feathers do look noisy. The elk portrait also hides the noise well in the facial fur.
All the above is good advice, but how practical is it? Far more important than stressing about the issues above is to get the shot. If you need high ISO to get the shot, then by all means use high ISO. If you need to underexpose most of the frame to protect highlights elsewhere, then do it. There’s a lot of post-processing trickery we can do to minimize the resulting noise so let’s get to that.
3) Minimizing Noise in Lightroom 6
There are tons of tricks to minimizing noise in Lightroom. I’ll go through some of my favorites. Be aware that every photo is an individual case and not every trick will work with every photo.
Most of these tips don’t utilize the noise reduction sliders. The luminance noise reduction slider blurs away detail to smooth out noise. I’m a stickler for retaining fine detail and I don’t like how heavy amounts of noise reduction make images look plasticky or like paintings. I only use the NR sliders after I’ve done everything else I can.
Because dealing with noise is so individualistic, don’t be afraid to work and re-work images until you get the results you like.
Let’s consider a couple of Lightroom basics. First, when you import RAW files into Lightroom, there is a default level of sharpening and color noise reduction applied. Second, Lightroom’s develop module is designed to work best from top to bottom on the right side panels. Hence I suggest doing your tone mapping (exposure, contrast and color tweaks) before dealing with the detail panel (sharpening and noise reduction).
When doing these tone mapping tweaks consider whether you want to apply them globally (to the entire image) or locally (to just a part of the image by using the adjustment brush). It’s tempting to just hit a global preset like the “Punch” option and be done with it. However, the tweaks you make to exposure, contrast and clarity can involve amplifying the signal and that of course increases noise.
Let’s work an example here. We’ll use our Burrowing Owl buddy.
Here I’ve taken the file and done the following global adjustments:
Exposure +1.00
Contrast +19
Shadows +21
White +30
Black -31
Clarity +31
Vibrance +11
Saturation +5
Sharpening 75, 0.8, 50
Noise reduction was at default: luminance 0, color 25.
The result looks good and punchy but full of noise. Check it out at 100%:
Blechhhh! Look at all that luminance noise. Well, I’d better apply some noise reduction.
I’ve added +50 luminance noise reduction and much of the noise is gone. Let’s push the slider to +75 and obliterate the rest.
Now the noise is gone, but so is the fine feather detail, especially in those cool snowflakey bristles around the bridge of the beak.
Consider if the entire image is benefiting from your tweaks or just a portion. If it’s just a portion, then go into the adjustment brush to make a local adjustment. Case in point is the Clarity slider, or as I call it the “Add Noise” slider. This slider boosts mid-range contrast while leaving highlights and darks alone – it also enhances the appearance of noise. Because of this I suggest that if noise is an issue with your image, then only use Clarity locally. I often do this with bird photos as clarity can really make feather detail pop and as demonstrated previously, the noise increase gets hidden by all that fine feather detail. If I were to apply the clarity globally, then the smooth background around the bird becomes noisy.
Here’s our owl again with the global tweaks and at 100%:
And now lets only adjust locally. Our first pass with local adjustments is to the owl’s face:
The adjustment brush was set at:
Contrast +100
Clarity +100
Sharpness +50
(Global adjustments are exposure +1.00, white +25, black -11, clarity 0, vibrance +11, saturation +0.5)
That’s a pretty aggressive adjustment brush setting so I set density at 100% but flow at 55 so I could gradually brush the adjustment on until it popped like the full global version, but just on the face. Now the bird’s face looks very similar to the first version, but the background noise looks finer grained because it hasn’t been tweaked.
My local adjustments to the face have left the eyes noisy so I go back to the adjustment brush and set it to shadows – 50, sharpness -74 and noise +76. I brush just the eyes and get this:
Another way to blur away noise is to use the adjustment brush with a negative clarity setting and paint away noisy sections. (Pushing clarity to the right increases mid-range contrast; pushing it left reduces mid-range contrast giving a softening effect.)
Here I’ve brushed a local adjustment of -70 clarity to the background.
In my experience this works best for big blank stretches of noisy background, this background is a tad busy for best results. Note that if you got the noise by globally increasing clarity that this will just reset the clarity amount for the area painted over. For example, if you increased clarity to boost cloud detail, but then paint over with a negative clarity adjustment brush you’ll lose that cloud detail you worked so hard to get.
We’ve still got some fine noise here. At this point I don’t find it objectionable and it will probably go away if I downsample this to print it at 8”x10” (this small owl will be life size at 8×10). But what if we want to print larger than life? Let’s go to the background adjustment and push the noise slider to +80 and get this:
Go back to the full frame crop for final output, in this case our 2048 pixel tall web output and here’s our result.
Pretty decent for a 10000 ISO crop sensor shot that was underexposed a full stop and we never even touched the noise reduction panel (though we did apply some local noise reduction to the eyes and background).
The other big popular “Add Noise” culprit is the Shadow slider. There seems to be a trendy obsession with showing off shadow detail in photos and I admit I fall for this a lot. However there is nothing wrong with shadows being shadows and hiding detail – this can add mystery to a photo. So before reflexively trying to recover all the shadow detail you can ask yourself does this particular image really need it? If yes, then get ready to deal with the noise revealed.
Here I’m messing around trying out a 1.4x teleconverter on a 800mm lens and due to the reach and some heat shimmers I’ve got some blurring so I’ve already applied some sharpening to this (78,0.9,51). I’ve exposed not to blow out the highlights on the rimlit wings. This is a decision I’ll question later, but for now I’m stuck with this file and there’s not much detail on the side of the wings we’re looking at. We’ll zoom in to about 170%. At 1600 ISO on the D810 it’s no surprise we’re seeing noise in the background. Let’s try and recover some detail in the wings by running the shadow slider to +100.
Yuck. Now we see a bunch of noise in the wings. Because the background is dark the noise is getting worse there too. I would have been smarter lightening the wings as a local adjustment, but for the sake of example let’s continue down the path we started on. I’ll apply some more tweaks as follows to get close to where I want the tonal values to be. I’ll deal with the noise after I get the tone mapping done. Adjustments:
Exposure +.80
Highlights -100
Shadows +100
White +32
Clarity +36
Vibrance +20
Saturation +16
Those adjustments to exposure, shadows and clarity aren’t helping my noise situation.
Time to head into the detail panel where we find the sharpening and noise reduction controls.
4) Sharpening
Sharpening images increases noise, however all RAW files need some sharpening to look presentable. In Lightroom all RAW files get the following sharpening defaults: amount 25, radius 1.0, detail 25. Sharpening works by detecting edges and increasing contrast along those edges and you guessed it, by increasing contrast it increases noise. Here’s where my favorite trick works wonders. Look at the masking slider – it has a default of 0. This means sharpening is being applied to the entire image. However by moving this slider we can mask away the parts of the image with less noticeable edges and only sharpen the more prominent edges. By using masking we can take this from a global adjustment to a local one. To judge what is being masked hold down the option key (or ALT key on Windows) while moving the masking slider. The screen will be all white at the 0 default. As we push the slider right, the image will gradually darken in the areas with less distinct edges. See Nasim’s excellent article on how to sharpen images in Lightroom to understand how this tool works in detail.
I’ve pushed the slider to about 25 and this is what I see. The areas in white are being sharpened; the areas in black are not getting sharpened. I’ll keep pushing the slider right until only the areas I want sharpened are still highlighted in white like this.
Voila – the bird is highlighted but the background is black, hence only the bird will get sharpened. Background noise will be minimized. In this case the white areas keep the 78,0.9,51 sharpening settings I already dialed in and the black areas get zero. Looking closely here’s the result.
In my opinion this is the single best Lightroom trick for keeping noise down. Here’s another example side-by-side at ~200%.
Unmasked.
Masked.
You can benefit from this trick even if you don’t add sharpening to your image because Lightroom will have already put its +25 default sharpening on your RAW file.
The Detail slider in the sharpening panel also boosts noise when pushed to the right. When pushed right it tries to analyze blur and sharpen to correct it. When your base level of noise is low, the added noise will be small, but if you are already dealing with a noisy file the noise will exponentially increase.
The radius slider doesn’t increase or decrease noise but can sometimes make the little “grains” look smaller (with smaller radius) or bigger (set at larger radius). The smaller grains may be easier to hide using other tricks later, but I recommend not sweating the radius setting for it’s effect on noise but instead utilize it to achieve the sharpening effect you want based on the spacing of the details in your image.
One other sharpening trick is to go to the adjustment brush and only sharpen the areas that really need it. Pushed to the right of center this increases the sharpening amount (like with the detail panel slider).
At last, we’ve finally made down the developing panels to the noise reduction sliders. But wait, there’s one super important thing we need to consider before messing with those. This is final output – i.e. what size and format the final image will be viewed at. If the image will be an 8×10-inch print on our wall it’s important we view it at that size. If it looks good at that size then we’re done. One huge mistake many of us make is that we obsessively view our images at 1:1. At that magnification any noise will jump out at us. But if the final output is to be viewed at less than 1:1 then our output file will be downsampled accordingly and downsampling does wonders for hiding noise.
This crop sensor ISO 7200 file is straight out of the camera with zero postprocessing other than the Lightroom defaults. You saw it at 100% at the beginning of the post and the noise was apparent. Now the noise is pretty much gone all from just downsampling the file to web resolution.
If we bust out a test print, evaluate it from a normal viewing distance, and still don’t like how much noise we see, then we might consider going to the NR sliders, but first consider if the file may have been overcropped and if a looser crop will still look good and shrink the noise away as well.
5) Noise Reduction in Lightroom
Still not happy? Time to blur that noise away. We can do that globally or locally. The simplest method is to go to the noise reduction panel. Default is set to 0 for luminance and 25 for color noise.
When noise gets really bad, such as in the opening fires of hell shot you will start to see horizontal banding artifacts or sometimes both horizontal and vertical forming a checkerboard pattern.
(Note: I increased exposure, shadows, clarity, contrast and sharpness in that output to make the noise worse so you can see the banding effect.)
The noise reduction sliders work by blurring the image – this can wreak havoc with details. If you add excessive noise reduction you get a result like this (100% view).
Hence it’s a real back-and-forth game adjusting sharpening and noise reduction to retain as much detail as you can (without halos or other artifacts) and keep the grainy look to a minimum. I find that by applying all the other noise minimization techniques we talked about above, that I rarely need to push the amount slider beyond 10. This while being primarily a wildlife photographer who shoots a lot at high ISO. I’d rather have a slightly grainy looking image than a smooth waxy-looking one – just a matter of taste.
The same 7200 ISO shot above, with added sharpening, masking the sharpening and +15 NR.
You can also use the noise setting in the adjustment brush to apply noise reduction locally. This adjustment works like the luminance amount slider but without detail and contrast adjustments. This can come in handy if you did some local adjustments, e.g. dodging, that caused noise in just that section of the image. You can actually go to that local adjustment, click on the dot marking it on the image and the setting panel will show the prior adjustments. You can then just add the noise adjustment to that set rather than paint a whole new mask.
Because I rarely go over 10 on the luminance amount the luminance detail and contrast sliders don’t make a significant difference so I just leave them at default.
Color noise is rarely apparent unless you’re shooting quintuple digit ISOs. If I see tell tale color noise at the default setting of 25, I will push the color slider right until the color blobs are minimized.
Here’s our color noise shot again. For this version I reduced the color noise correction from the default setting of 25 to zero to show what color noise looks like.
Here we see the color noise gone when we just accept the default amount of 25. Unless you deal with a bunch of night sky photos, chances are you will never need to monkey with the color noise sliders.
The color noise detail and smoothness sliders might be of use in images with extreme color noise, but I can count the number of times I’ve used these on my fingers. In theory they can make colors bleed and look smeary or desaturated along edges. I reckon if I had an important image with noise so bad I needed these that I’d try a different software program such as DeNoise first.
6) Summary
Dealing with noise has a lot more to do with noise avoidance than noise reduction. Avoiding noise in camera and during tone mapping is the best strategy. By doing this I find I rarely need to push the noise reduction sliders. The amount of noise reduction needed depends entirely on final output. Hence noise reduction should only be done after final output is decided. And when all else fails you can always do the gritty black-and-white conversion trick and your friends will think you’re such a great artist they’ll all give you black turtlenecks for Christmas.
P.S. Nasim has previously published a detailed tutorial on noise reduction that I recommend you take a look at, where he utilizes external tools to reduce noise. Selective noise reduction can yield excellent results when using such tools as Nik Dfine.
All content and photos ©John Sherman unless otherwise attributed. No reproduction without prior written permission.
Very helpful. Your great practical tips and step-by-step examples/settings will be most useful to one who has an unhealthy aversion to noise. Thanks for taking the time and effort in putting this tutorial together.
Thank you so much,sir.
Good article, John.
One big factor with noise is that most photographers create noise or make it worse early in their workflow. Are you adding contrast with a preset when you import using ACR or Lightroom? Are you using sharpening with an ingest preset? Both of these routine steps can add noise to a problem image. When you have a problem image – such as high noise – create as neutral a RAW conversion as possible and avoid applying contrast or sharpening at ingest or import. Make your edits then apply contrast and sharpening as needed – often selectively.
Even worse, a Landscape preset that adds contrast and makes shadows darker – but you need to brighten the shadows and that adds noise. A neutral preset helps to avoid creating noise in shadows when you brighten those dark areas later.
Great point about the presets Eric. I’m all about exercising noise avoidance instead of applying noise reduction and your observation adds to this message. Thanks.
Thank you so much, John! This was the best article I’ve read in a while (while they are all good, Nasim). I learned something about sharpening that I never learned before with Lightroom, the alt-key trick!
Thank you very much John for a brilliant article. I learnt a lot from this article and very nice of you to use examples from the field, as it makes relating to the noise avoidance and reduction much much easier.
A minor comment: some cameras (those that use Aptina’s DR Pix technology for their sensors, like recent SONY cameras do) effectively have two “base ISO” values.
SONY A7RMII, for example, has those at 100 and 640. Noise-wise, it makes sense to shoot at ISO 100 – 200, and after that to go to ISO 640, skipping all the intermediate values; and up to ISO 6400. Read noise at ISO 400, for example, is nearly 2x larger than read noise at ISO 640; while read noise at ISO 200 is about the same as read noise at ISO 640. Read noise at ISO 500 is about the same as read noise at ISO 1600. After 6400 ISO becomes “digital”. Also, “numbered” values of ISO 50, 64, and 80 are fake, they are the same as 100, 125, and 160.
You can see it on the graph Alex Tutubalin made: http://blog.lexa.ru/sites/blog.lexa.ru/files/a7r2-black-noise.png
Is this similar to the theory that Canon cameras’ “native ISO” or “true ISO” settings are the analog multiples of 160 (ISO 160, ISO 320, ISO 640, etc), and that all other ISO settings in-between are produced digitally and therefore noisier as a result of digital push? Proponents believe ISO 1250 can look on par with or even better than ISO 125 on that basis.
No, it is not the same. You may want to read Aptina’s white paper on DR Pix.
Good job John!
You left nothing to be explained and was thorough and clear in your tutorial.
What a great tutorial! I just recently purchased a Nikon D750 upgraded to Lightroom 6, and am trying to settle in to a workflow that is more disciplined and simplistic. This article is just what I needed! Thank you.
Hi John, a great explanation and wasted on me as I 1. don’t posess Lightroom and 2. I use DXO Optics Pro 10 Elite for this very problem. It has an amazing and new ‘PRIME’ noise reduction slider, which I will come to in a moment. The great thing about the software is that it has a first attack on my RAW files with a default ‘standard’ setting, which tackles barrel / pincushion distortion on a lens-by-lens (with a specific camera body) basis.
The vendors, DXO Mark, run all popular and some not so popular lenses, in conjunction with camera bodies, through their lab, and produce a module for each profile, which auto downloads when an image created with a new combination is detected. For example, if I went out and bought a D810 today, and popped my Sigma 105mm macro on it, and took some images, if I then loaded the resulting RAW into DXO Optics Pro, it would detect the new combination and present me with a pop up window ‘New Profile detected, would you like to download a new profile for Nikon D810 and Sigma 105mm Macro Lens? Y/N.
Hit Y, and the profile is downloaded, and is very specific to you own camera, and your own lens. From there, the standard corrections are dealt with, colour balance, distortion, etc etc. Noise is NOT dealt with, there is a slider for it, with a choice of standard or prime. Prime takes a little time (on my old computer 40 seconds, but on my new i7, just 3-4 seconds) and when done, gives an amazing result and is quite effortless.
By the way, if you don’t want the program to do a default ‘standard’ correction, there is a ‘no correction’ profile you can run, takes a second, and gets you back to the unadjusted RAW, but the ‘standard’ correction is pretty close to what I would have done by hand anyway, so in most cases, I let this stand, and then make any fine tuning, plus noise reduction, myself.
Hope this helps, it is a lazy solution but, it does work, if you don’t want to spend hours in post processing.
Of course, with an outstanding image, for a competition, or an important client, you may choose to Lightroom it instead.
Optics Pro should be seen as a RAW processor and converter, in the workflow, I often end up in Photoshop for localised correction, as optics pro has no area selection, and applies changes to the whole image, so it is a part of the workflow but, in many cases, I don’t need Photoshop (camera raw) anyway.
Hope this helps, great article, but for us lesser mortals, I thought I should mention this software too.
Came here to say something along these lines. Have an upvote :)
Another up-vote. Any time I’m handling files shot at ISO 800 or above I leave Lightroom behind and go to DxO Optics Pro. It’s well ahead in both noise reduction and colour fidelity on high ISO raw files.
DxO’s Prime has no secret noise reduction magic, it basically blurs details and saps colors like every other noise reduction plug-in, just taking minutes to do what other programs do in seconds.
Go to Dpreview’s retouching forum and you’ll see Prime routinely being beaten in a number of noise comparison tests even on ultra-high ISO RAW’s from Sony’s A7s. When you deduct points for DxO’s worse highlight recovery and the numerous digital edge artifacts it generates from its overly aggressive demosaicing engine it often ends up on the bottom of RAW convertors for overall image quality.
Hi Marco, your points are well noted. I would point out though, that Optics Pro could be thought of as a RAW converter with lots of options included. The Prime noise feature has been very well received by many online comments and reviews that I have seen, but I confess not to having visited the Dpreview site that you mentioned.
Was this a recent review? I know prime first appeared in version 9 of Optics Pro, and I am running the relatively new v.10 Elite, and I can assure you there is no issue with highlight recovery, or the other issues you mentioned relating to the Raw files from a Sony A7.
My Nikon does not seem to upset Optics Pro in the way that you say a Sony A7 does, so it’s just as well I stuck with Nikon.
As in any Camera Raw software, the controls on Optics Pro are all sliders, so maybe the reviewer was hitting the Sony images with a full blast, which of course you would not normally do. Have you tried Optics Pro yourself?
I KNOW for a fact that the Prime Noise feature in it far surpasses the Camera Raw I have with Photoshop 6. I have tried at least a dozen images using both products and I can never get anywhere near the results that Optics Pro produces, when I try to use Photoshop 6 Camera Raw (v 8.1 if I recall correctly).
Maybe Optics Pro performs better on the more popular camera brands (ok, I mean Nikon and Canon) than anything else?
No complaining here though, I am very pleased despite being forced to upgrade due to a lack of camera support on V.9, which annoyed me, so my view is hard won by Optics Pro, they are not my best friends at the moment, but I am at least, very happy with the product so far…. but now you’ve got me worried!
Hi Ross, check out these comparisons on Sony and Olympus RAW’s to get a flavor of what I’m describing:
http://www.dpreview.com/forums/post/54955221
http://www.dpreview.com/forums/post/56472532
http://www.dpreview.com/forums/post/56405302
I’m also a Nikon shooter like you and my own experience with my high ISO Nikon RAW’s basically mirrors what the Dpreview posters have reported on these Sony and Olympus files. In artificial test situations involving flat tones and high contrast edges (like text on paper, printed circuit boards, etc.) you can sometimes get Prime to produce cleaner noise reduction than many other comparative methods. But for real world images of people, landscapes, and other tonally complex subjects the noise reduction offered by Photo Ninja looks better than the output of Prime. Plus, you get Photo Ninja’s other image quality benefits… artifact-free demosaicing, best-of-class highlight recovery, and industry-leading CA removal.
Ultimately, we all use the tools we either have or feel most comfortable with so if you feel you are getting what you need out of DxO, great. The only point I wanted to make in my post is that almost all noise reduction software is roughly equivalent these days. You’ll also notice that’s why there’s been no further meaningful development in this field for the last several years. As such I’d choose a workflow or RAW convertor not for noise reduction but rather for its output image quality with regard to color, tone, and demosaicing. YMMV
Hi Marco, sorry for delayed reply, got a huge project on here and my time is very stretched just now.
I had a look at those links you posted above, thanks for that. I misunderstood your original reply to me, when you mentioned the DPreview.com test, I took you to mean that the website itself had conducted a full scientific test on the software, but the links point to DPreview forum posts where individuals are reporting their own views based on their images and processing.
I glean from this, after reading through them, that the superior processing given by the Ninja software, over DXO-Prime was based on quite high, to very high, ISO settings. On one of the postings there are images of a cathedrals’ stained glass windows, and though the Ninja results do, at first glance, seem better, I felt, after viewing them a few times, they were a little over-saturated, but I guess this could be adjusted in the Ninja software.
Just to put you right on my own thinking, I tend not to use anything higher than iso 400 because I have invested a fair bit of money in both equipment (camera, lenses & now a new computer) and software, and to secure the best outcome, keeping the iso low is the right thing to do. Clearly, shooting at extra high iso settings is a recipe for noise ridden images, and the need for software that is perhaps designed for that class of image, where the noise is so bad that the image has to be rescued.
This is where perhaps we differ in our shooting. I find that, when pixcel peeping, even on relatively low iso settings, there are certain kinds of images, chiefly those with a high dynamic range, that produce some noise and chromatic abberation problems, in the shadow and highlight areas of the image, respectively. I can vouch for DXO Prime noise removal in such cases, and their Optics Pro deals with the chromatic abberation extremely well too.
I shoot mainly landscape and macro, and often use a tripod, so you’ll see that I am not going to be using high iso settings for such work anyway.
It could well be that the Ninja software is better for such high iso imagery, I think anyone going to be shooting often in that kind of mode should be looking at all the noise reduction software available to better the results, but for those who are shooting at conservative iso settings, looking for the best quality image possible, I am very happy with what DXO Prime is delivering.
Look at some of the articles on here by Thomas Stirr, I know he is a big fan of the Optics Pro software, and he shoots a lot of wildlife and landscape.
I think the lesson for us all is to try to keep away from high iso settings if at all possible, just because a camera ‘can’ shoot up there, does not attract me to risk my images if a lower iso can be employed. It is the old ‘if it ‘aint broke, don’t fix it’ argument – sorry :)
Hi Ross, agreed… the best offense is a strong defense. Keeping your ISO low trumps any noise reduction method hands down.
The Dpreview examples were obviously focused on high ISO noise reduction because I think this is where the posters felt firmer distinctions in noise handling between RAW convertors could be found. In low ISO noise reduction usage it’s quite conceivable that there’d be hardly any meaningful difference in performance.
This brings us full circle to my earlier point. All of us are best served by the RAW convertor that handles the other image quality aspects of the conversion best for our particular cameras (demosaicing, color, tone, lens correction, UI, etc).
I’ve personally compared DxO v8 on a number of my Nikon NEF’s, for example, and consistently found demosaicing artifacts and jagged edges when comparing images 1:1 against conversions done by Photo Ninja and Capture One. Since DxO doesn’t allow me to use alternative demosaicing engines or tinker with its overly aggressive demosaicing settings, my conclusion can only be that DxO clearly wasn’t the optimal convertor for me.
I’d encourage you to download either the easily available trials for Capture One or Photo Ninja and run some comparisons of your against DxO with your own NEF’s. At a minimum you can confirm that your DxO conversions can’t be beat. But quite possibly you might also be pleasantly surprised that you can elevate your low ISO RAW conversions yet another notch!
Hi Marco, I’ll certainly give the software you mention a try. I don’t think we should assume that all noise reduction software will be similar when applied to relatively low iso range images, it could well be that Optics Pro is actually better at low to medium iso noise issues, whilst those you mention are designed for high iso nois issues, more specifically – who knows?
One thing caught my eye in your last reply, it was the reference to using version 8 personally to compare your nikon NEF files (against other noise reduction software etc). I would have to point out, for anyone else following this thread, that Optics Pro didn’t introduce PRIME noise reduction until version 9.0, and have considerably improved it in version 10 Elite, the present version. In fact, I got hold of version 8 free on a magazine cover here in the UK approx six months ago, and found it awful. Truth be told, if I had not discovered the software initially, when it was already up to version 9 (the first time I bought and tried it) I would probably have been looking elsewhere.
Anyone can download a free copy of version 10 to try for a month, after which, you have to purchase or cease using, so if anyone out there is curious, give it a go! The interface is awful… but the results are worth the learning curve imho.
Regards, Ross
Excellent. Thank You !
Great Explaination!
May it is possible to present before/After pictures with one of these sliders ( you move the sliders from left to right to see the effects).
Some of the examples are at a so high level/quality ist difficult to See the Difference while scrolling the screen…
Nice Greets from austria!
Hello Wolfgang,
I agree it can be hard to see the subtle difference at web resolution. I think the best way to deal with this is for readers to take a high noise photo of their own and apply these techniques to it. Not all these tricks will work with every photo and these tricks only came to me after years of experimenting. Like I said, every photo is different and the approach to each one should be individualized. If I hadn’t played with these techniques on thousands of photos I would not have been able to write the post so I encourage everyone wanting to improve their post processing to practice, practice, practice. I hope the info I shared can make your learning curve go smoother than mine.
Verm
I often find in bird photography I have to choose between a less-sharp aperture (5.6) with an acceptable ISO (500), or a sharp aperture (8.0) with a noisy ISO (1000). Which of these is more appealing to you?
Thanks.
I will always opt of the exposure settings I think are best. Like I said in the article, noise is the last nitpicky thing I worry about when it comes to capturing my images. I shoot at f/8 a lot more than I do wide open because DOF on supreteles is so shallow. If f/8 gives me 3 inches more of DOF that can make or break a photo. GET THE SHOT, deal with the noise later.
Very interesting article. I couldn´t agree more. The plastic look of excess noise reduction is atrocious. I try to get my pictures correctly exposed in the first place and accept that sometimes it is not posible to make a decent picture, even with modern sensors and clever image cooking software.
What makes you think you know anything about noise, John? Huh? I never see noise in any of your photos.
Oh, wait. After reading your article, now I know why. No wonder muting my in-camera microphone wasn’t doing the job.
Thanks for the informative collection of words and photos. I’ve learned something here.
Well written and very practical article – one of the most helpful I have read in months. Thanks.
Good article. When shooting a lot of pics some of this stuff is simply too much work. It dawned on my when I was reading this that it would be great if LR had a reverse masking slider for noise reduction, so you can apply more NR to low frequency areas.
Amazed at how shallow the DOF is at 600mm f/10… :-O
Those were shot at close to minimum focusing distance (Burrowing Owls are pretty cooperative, hence why you see so many good shots of them) and you’re right, DOF at 10 to 15 feet away is pretty slim at 600mm.
John,
Thank you for taking the time to write such a great article.
Don
Thank you for a much needed and appreciated article.
I’m wondering if you are saying that my indoor sports photos in general would have less noise if I shot at +1 with a 16000 ISO and lowering the exposure in LR as compared to shooting with 0 camera exposure compensation and 8000 ISO while not having to make any exposure adjustment in LR? Or is that more for higher detailed photos?
I would stick at ISO 8000 and proper exposure. ETTR works best at low ISOs and dialing everything in perfectly. With your sports shots, capturing the perfect moment is way more important that noise. If you are shooting lots of indoor sports and want to recover shadow detail, say the face inside a helmet, then owning a fat pixel camera like the D4s is the way to go. Let emotion and excitement carry your sports photos and nobody will notice the noise.
All that said, I would do some experiments on my own in the conditions I shoot in (I do this all the time) to see if I like the results better one way than the other.
Verm
ISO is not a part of the exposure, and exposing properly while having ISO at 8000 you risk severely and unnecessarily blowing highlights off. No camera actually benefits from ISO 8000, too. The highest mark among non-MF cameras goes to SONY ILCE-7RM2, it benefits from ISO bumping up to ISO 5800. For D4s and Df it is less, 4800. Very few cameras go up to 4000.
I will certainly defer to Iliah on technical matters regarding sensors and how they record data (just read some of his posts on PL and you’ll know why). Iliah writes photo software and can out techgeek Nasim so I hope he can help us answer Bill’s question when it comes to what camera settings to use.
I’ll paraphrase what I think the practical consideration is: if I was shooting an indoor sports event and my minimum exposure requirement to get a sharp photo was 1/500 sec wide open at f/4 and to achieve this my meter said I should shoot at 8000 ISO, would I be better to shoot my D4s at 4800 ISO, 1/500, f/4 then try to achieve the desired brightness by adjusting exposure in post? Or is it better to amplify the signal in camera and do as little amplification in post as possible? Or does this vary from camera to camera?
Also I can understand that if one defines exposure as how much light hits the sensor — i.e. how much comes through the given aperture (f-stop) and how long it hits the sensor (shutter speed) — that it is correct to say ISO is not part of exposure, but instead just a degree of amplification applied to that exposure. However the concept of the exposure triangle (shutter speed, f-stop and ISO) is prevalent in photo teaching today, so I hope you can clarify what you mean by ISO not being part of exposure.
Regards,
Verm
> if I was shooting an indoor sports event and my minimum exposure requirement to get a sharp photo was 1/500 sec wide open at f/4 and to achieve this my meter said I should shoot at 8000 ISO, would I be better to shoot my D4s at 4800 ISO, 1/500, f/4 then try to achieve the desired brightness by adjusting exposure in post?
Yes, there is no point in going beyond 4800 with D4s, all one achieves when raising ISO above 4800 is more clipping, noise-wise nothing is achieved. “Adjusting exposure in post” may be a confusing term, let’s say “adjusting brightness”. With many converters “exposure” slider is not what one is looking for to adjust brightness while preserving highlights, curves often serve better.
The point where the raise of ISO does not result in lowering noise depends on the camera, and somewhat on the type of light.
> the concept of the exposure triangle (shutter speed, f-stop and ISO) is prevalent in photo teaching today
Too bad, especially when it says “the higher is ISO, the higher is the noise” – too often the exact opposite is the case.
Thank you Verm and Iliah. So I should be setting my Auto ISO to 5000 max, while shooting 1/800 and f2.8 for winter indoor sports with 0 exposure compensation, aperture and shutter locked and than adjust the brightness in LR or DXO? Never realized a correlation between the higher ISO and any clipping and I’ve been running my D4S Auto ISO maxed at 25,600 even for Friday night football games where it’ll run between 8,000 and 12,800 ISO and a D4 before that. So the camera is just adding brightness to the RAW files when the ISO is above 4800 being somewhat careful to avoid a lot of clipping?
Bill,
I would always practice new settings and new post-processing techniques before putting them into practice on an important shoot. I’m eager to do some testing on my own based on the info Iliah has shared.
Verm
Iliah, where are you getting these max ISO figures for the ILCE-7RM2 and D4s? If the source is a publicly available table, I’m interested to know what the same respective max beneficial ISO values would be for my D200 and D700.
From my experiments. Bill’s measurements indicate that for D4S the improvement in shadows past 1600 is less than 1/6 EV; and the same is true for A7RM2 after ISO 640. See http://www.photonstophotos.net/Charts/PDR_Shadow.htm
Verm,
Luminous exposure (Hv) is expressed in units of lux seconds. ISO sensitivity is a standardized means of expressing the relationship between Hv and the resulting “brightness” of a recording medium/device.
We can deduce from the above that the “exposure slider” is incorrectly named: it is impossible to change Hv either retrospectively or by changing the ISO sensitivity of the recording medium. Hv is determined entirely by the scene illuminance, the effective T-stop, and the shutter speed that were used to record the scene.
At high values of Hv there are ample photons recorded by the medium for the effect of photon shot noise to be invisible in the image. Photon signal-to-noise ratio (SNR) = the square root of the number of photons recorded. E.g. if a pixel records 10,000 photons the photon SNR will be 100.
At low values of Hv, such as photographing action in dim light, let’s say we have a maximum of 100 photons captured per pixel. The photon SNR is now only 10 at best, which will produce easily visible noise in the image. To achieve the same image brightness as my previous example, we need to multiply the photon counts by a factor of 100. If the previous example was using ISO 100, this example requires ISO 10,000.
Intuitively, we assume that setting our camera ISO to 10,000 (or selecting auto ISO) would be correct for my second example. It would be correct only when we are shooting JPEGs and not RAW. As Iliah has pointed out, when shooting RAW the optimum ISO value to set in the camera is far from obvious, and it varies greatly between makes and models of camera.
There are several factors that combine to cause this anomaly. One of the main factors is, I think, users expecting cameras to provide ISO settings that are adjustable in steps of less than whole f-stops. Let’s suppose we use our camera at base ISO plus one third of an f-stop: this requires that the photon counts are multiplied by the cube root of 2, approx. 1.25992. Because the multiplied counts have to be expressed as integers in the RAW file, it results in severe non-linearity of the low count levels: 1 produces 1*1.0; 2 produces 3 = 2*1.5; 3*1.333; 4*1.25; 5*1.2; 6*1.333; 7*1.143; 8*1.25; 9*1.222; 10*1.3; 11*1.273; etc. These results are far away from our wanted and expected ISO value that we have, or our camera has, selected for our photograph!
The net effect of this non-linearity is a reduced signal-to-noise ratio in the shadow areas. The very areas in which both photon shot noise and sensor read noise are the most noticeable, and the most difficult to remove in software without smearing the spatial resolution of the camera system. Multiplying, in camera, the raw photon counts by only integer powers of 2 would seem to be the obvious way to avoid this particular problem. However, Iliah has clearly shown us that there are other factors involved in the design of various cameras that cause our seemingly obvious deductions / rules of thumb to not work out for the best in practice.
Many thanks for you informative (and great fun to read) article and your subsequent comments,
Pete
Nicely put! Thanks for putting this is in layman’s terms (or as close as one can get to those in this technical a subject).
Thanks for writing such an informative article. I learned a lot.
Thanks John .
When I used capture NX2 I used to use the control points for local adjusments and it worked wonder that tool was fast and amazing! Now that I made the move to the better software lightroomn I see my self doing less and less local adjustments because it is very time consuming and needs precision with the brush and you easily overlap the desired ”sharpenned area” . I just wish the u-points/control points with opacity would be bought by lightroom since NIK doesnt seem to be doin anything new with it anyways . And also using nik has a second software is also too much time consuming for me most of the time! ! !!
Thanks again for the article I’ve learned and ill be easy on clarity & shadows recovery ;) !!
You’re my hero…
Thanks for all of the detailed commentary; it’s clear you put an impressive amount of time and effort into your articles and love what you do. I have a D7200 and have been somewhat disappointed with the amount of noise I’m seeing in my RAW files compared to the D5200 I used previously.
I understand that there are no hard and fast rules and that everything depends on the image, but do you have any recommendations for what you typically find necessary as far as sharpening values or noise reduction values for the D7200 as RAW files have no sharpening applied out of camera? I typically aim to shoot at ISO values of 3200 or less (ideally 1600). I read on another website that recommended noise reduction settings of 25 for ISOs 100-400 and 50 for ISOs 400-1600. This seems excessive in my limited experience.
I would greatly appreciate your guidance. Thanks!
The recommended settings are zero noise reduction if possible. I would take a few noisy sample photos of yours and practice the techniques in this article. I think you will find you will need to use very little noise reduction after doing noise avoidance. I would never apply a default luminance noise reduction to any file, however I find no problem sticking with the default 25 for color noise reduction.
I’m curious at what step in the post editing do you correct the image for noise? At the end of start or towards the end just before you sharpen the images?
Thanks
Duane
Extremely enlightening, I am suffering on my d810 from noise at ISO200, on long exposure with a 24mm F1.4 at 2sec. So with these new knowledge, I will resett the all the pertaining in camera adjustments to try to rectify the problem. That even Nikon for professional users haven’t been able to help me with.
Hi from Brazil. The best avoid and reduce noise article ever! Congratulations!
Evandro
No new information given in this article. First it starts with general tips, how to quickly reduce noise in images. Then, there’s the great new tip on best shooting at base iso and to avoid high iso settings – wow, really useful information. And finally, the article turns into creatings masks in a graphics program, so that you can blur the background in the darker areas and keep the noisy, sharp details in areas, that are brighter and show more details and thus, better hide the noise from the eye of the viewer, cause you always have to play with the inflexibility of the eye to differentiate the details, that is more attracted to colors, form, contrast and details in brighter areas and especially faces and emotions like eyes or mouth. Thats nothing new. It won’t help the people to go into heavy manual processing, because you won’t generally spend hundreds of hours in front of thousands of images, that were taken just within 1 week. It’s just not economical. Hours of work are much more expensive than buying a full format sensor with moderate resolution and run it smoothly with F2.8 even lenses, maybe just stopped down to F4 to have some controllable DOF and to run the lenses in their sharpest aperture. Full format sensors run smoothly up to ISO6400, exactly the much needed range for fast wildlife pics and fast enough night shots below 1/125s, to freeze movements without relying on VR, that doesn’t make sense with moving subjects. APS-C is totally noisy at ISO3200, so even F2.8 lenses will be to slow, to keep ISO settings below ISO3200 to get minimum 1/125s shots for slow moving subjects in bad light. It would be better to give values in tables for the corresponding sensor generation within the range of particular camera bodies of each manufacturer – e.g. use HQ NR with 25-0 (NR amount-details) in software program X for ISO values 100-800, for ISO values 800-1600 use HQ NR 50-30, for values between 1600-3200 use this and that setting… The result woud be a slightly blurry image, but afterwards in conjunction with lowering the resolution from e.g. 24MP APS-C sensor down to 14MP 4583px width images 3:2 – the result would be clean and detailed enough, getting rid of the need of time spent in post processing for a picture, that is not even needed at 1:1 resolution, because a buyer needs the whole image, not a highly detailed crop out of it. If the pixels move together, the noise goes down significantly – nothing more than a big sensor with low resolution. Creating masks only helps sometimes, if you absolutely need the full resolution – but once again, no professional likes to spent hours in front of a screen, just to get rid of noise and getting lack of detail or to remove noise partially, thus having to think about which parts need to get destroyed or not. They just buy the biggest portable acquirable sensor, to avoid lack of detail and hours of post processing. They just want to tweak the images slightly, e.g. color adjustments, slight exposure changes. Every change of an image in a program will not increase image quality. The input must be perfect and then it gets worse and worse.