I was experimenting by taking photos of the moon using an 1000mm MTO-11ca lens and thought I should get even closer by adding a teleconverter (Kenko Pro 300 3x). The picture quality is not sharp at all but by using a realy high ISO (25000) and fast shutterspeed 1/2500 I was able to get rid of the worst blur (I was shooting on a tripod but I think atmospheric effects was moving the moon in and out of focus).

As could be expected the images where VERY noisy Links bellow. I usually only go up to ISO 1600 reluctantly (Canon 5D Mark II).

25% crop

Full size JPG Full size RAW

How much can be salvaged with images this noisy? I was trying both Digikam and RawTherapee but did not find any denoising algorithm that made any difference at all. I would have expected the denoising to at least smear out the picture and kill the noise even if it failed to recover the details.

What is the best that can be done in these cases?

Edit: Some more experiments have revealed that the ”athmospheric effect” might in reality be caused by my lens not being in temperature equlibrium as per this article https://fotosaurier.de/?p=2029

  • Atmospheric effects or tripod movement? at 3000mm, most tripods, particularly consumer grade ones, are not up to the task.
    – Michael C
    Feb 3, 2020 at 8:13
  • @Michael C, could be either. There was no wind and problem persisteed with 10 second timer so I guessed atmospheric, I thought that the moon going in and out of focus in the viewfinder and also in liveview should not happen as any small movement should not affect infinity focus. It looked more like a focus miss than camera shake to me but I might be mistaken. (Could also be play in the lens mount affecting the focus)
    – lijat
    Feb 3, 2020 at 8:47
  • 1
    If you have taken many different images, have you considered image stacking to reduce the noise?
    – Bob__
    Feb 4, 2020 at 8:57
  • @Bob, I have considered it but I think they are to different, I had problems with atmospheric distortion and focusing
    – lijat
    Feb 4, 2020 at 11:40
  • 2

9 Answers 9


Diffraction and noise

You basically used a 3000mm f/30 lens, so heavy diffraction is expected. You simply cannot obtain sharp results with this lens and teleconverter, even if you manage to focus the lens properly.

On top of that, the lens itself might be soft. With the teleconverter, you won't get any sharp result, without, you might.

Finally, you have heavy noise and banding, so this picture is pretty much a lost cause.

Possible post-processing

You could:

  • accept the noise and sharpen a lot, in order to at least see some details on the craters.
  • remove chroma noise by converting the picture to black and white
  • replace the background with pure black or a very dark grey in order to remove the most apparent banding.


enter image description here

Black & white picture. Clarity and sharpening set to 11 (in Lightroom):

enter image description here

Replacing the background with a very dark grey (e.g. with Lightroom or Photoshop). Quick and dirty job just to show the theory:

enter image description here

Shoot again

You'll have to shoot again, possibly without the teleconverter, with a much slower shutter-speed and much lower ISO. See looney 11 rule.

I have a 900mm f/8 telescope at home, and the specs are not too different from your lens. Here is a picture I shot a few years ago with a Fuji X100s, at f/5.6, 1/250s, ISO 800, simply shooting through the eyepiece:

enter image description here

Note that a half-moon would have had more contrast than a full moon, especially at the terminator.

  • What software are you using? Lightroom?
    – lijat
    Feb 3, 2020 at 23:00
  • 1
    @lijat lightroom for the first adjustment, photoshop for the background (lightroom could have been used too) Feb 4, 2020 at 5:45
  • 1
    Mmm, nice pic!!
    – Strawberry
    Feb 4, 2020 at 17:18

The moon is really a special case, because it is mostly grey. So you can remove chroma noise by just picking a channel (green, usually). This also removes some of the chromatic aberrations on the edges (when they are sharp, which isn't he case here).

You can also average the three color channels: copy the image to obtain three layers, and using the channel mixer, make a layer be (R,R,R), another one (G,G,G) and one (B,B,B). Then average them by setting the opacity of the top one to 33.3% and the next one to 50% (bottom stays at 100%).

enter image description here

If you want to restore the colors, you can blur the initial image and re-apply it in color mode over the clean image.

I also tried a median filter but the results are a bit fuzzier.

IMHO you have completely overdone it. The moon moves by its diameter in 120s. In your picture the diameter is about 4200px so it takes 1/35s to move by one pixel. 1/100s would give you a huge safety margin, and would still let you use much more reasonable ISOs.

  • The reason for the high shutter speed is not moon movement, there was some wierd atmospheric effect, i guess, so the moon moved in and out of focus and I had to hold the camera, on a tripod, and time when I took the picture. This nessitated a fast shutterspeed to overcome camera shake at 3000mm.
    – lijat
    Feb 3, 2020 at 7:16
  • If you're touching the camera at all at 3000mm, you're dealing with camera motion blur. 3000mm is not 30mm or even 300mm in that respect.
    – Michael C
    Feb 3, 2020 at 8:14
  • @Michael C, that was one of the reasons I went as high as 1/2500, my earlier shots using a timer and slower shutterspeeds 1/640 was worse in sharpness (but less noise due to a sane ISO).
    – lijat
    Feb 3, 2020 at 8:49
  • 3
    At that zoom, just the mirror moving might be causing motion blur... On a different note, if you're going to throw out colors, you'll probably do better to use a debayering process that does that, since potentially you can recover information from every pixel. Throwing out a channel after the fact means you've essentially ignored up to three quarters of the pixels that were exposed. (If you keep green, probably only half, but still...)
    – Matthew
    Feb 3, 2020 at 20:20

I think your image would have turned out much better without the 3x teleconverter. With a 18MP or 24MP camera, cropping the image would probably be sharper and you could have used a more reasonable ISO.

I use a free program by Imagenomic called Noiseware and here is a sample of how it works on a jpeg.

enter image description here

  • It did not, I was trying that and the results are marginally worse, not a whole lot different thou. It was being unsatisfied with cropping in that made me try this combo.
    – lijat
    Feb 3, 2020 at 7:12
  • That program looks interesting, will have to try that.
    – lijat
    Feb 3, 2020 at 7:18

As noted in other answers there, is neither going to be a single fix nor a complete set of fixes. It may be worth seeing what the DXO Photolab prime de-noising algorithm can do; it can be be extremely effective. My concern is the possibility of excess smoothing however it may be worth installing a trial to see if you can create a base onto which you can layer, with greater effectiveness, other adjustments (increased exposure and clarity will be your friends in my opinion).

  • My answer seems to have been sandwiched between two pre-existing answers. I don't know how this happened as I simply used the response box at the bottom of the page. In any case apologies and this was not intentional.
    – Adam Gold
    Feb 3, 2020 at 19:09
  • 2
    By default, answers get sorted by score, while answers with equal scores get sorted randomly.
    – Mark
    Feb 3, 2020 at 23:15

In general you can always remove noise, but you might not have much detail left in your image if the signal to noise ratio was poor to start with.

State of the art these days are filters based on a non-local means algorithm. This tends to do better than most previous algorithms in terms of not making textures look like smooth plastic, if there is some detail with a scale larger than the noise.

I mostly deal with video, not still images; FFmpeg includes a slow version of Dirk Farin's nl-means implmentation (most of the SSE SIMD optimization is disabled for some stupid reason in the FFmpeg version, making it nearly unusably slow for high-rez video, like multiple seconds per frame on i7-6700k Skylake at 4GHz at 1080p.) A more efficient version of the filter is available in HandBrake. But even that's too slow for realtime (24fps) on current CPUs; NL-means is very computationally expensive.

For a single still image, only spatial smoothing is relevant; no temporal (between frames of a video) smoothing is possible, of course, unless you have multiple shots of the exact same scene to average noise across them.

You can use ffmpeg on a single still image. e.g. ffmpeg -i input.png -vf nlmeans output.png. See the FFmpeg filters docs for it if you want to use it that way, for info on options you can use with -vf nlmeans=....

The same ffmpeg doc describes various other image noise-reduction filters that FFmpeg has.

  • Interestingly, although this algorithm works great for some images (I got some nice results for my other noisy image, using Wolfram Mathematica's implementation named NonlocalMeansFilter), it does almost nothing to the noise in the demosaiced raw photo from the OP.
    – Ruslan
    Feb 4, 2020 at 20:25
  • @Ruslan: Maybe with more aggressive settings since the noise is as strong as much of the detail? Or if the noise is stronger than the detail, maybe nlmeans won't get rid of it? In general you can always smooth away noise, e.g. with other filters (like ffmpeg's hqdn3d, pp7, or fspp) or with downscale with bilinear or bicubic interpolation (not a sharpening interpolation like lanczos). But usually that will destroy a lot of small-scale detail, and most people prefer noisy / grainy images to smooth plastic. (Noise gives the visual impression of detail even if there isn't much real detail) Feb 5, 2020 at 1:09

Here is my process in RawTherapee:

We start in the Raw tab. The first thing I notice is that you have horizontal banding. RawTherapee can deal with this using the "Line noise filter". Set the direction to "horizontal" and the value to something that seems to work (around 400). Note: this is not perfect, there is some residual banding. I think you'd need to take some darks to get rid of it completely.

The next step is to adjust the demosaicing. For noisy images, IGV is probably the best option, pumping "False color suppression steps" to 5.

Moving on to the noise tab, you can start with "Impulse noise reduction", which will take care of the "salt and pepper" noise. Then, on "Noise reduction", set the colour space to L*a*b, since you want it to work in luminance, and the mode to Aggressive, since there is plenty of noise to go around. Adjust the luminance slider to taste. You can also turn on the median filter, that will clean up a lot of what it is left. In my opinion, the best is the window of 5x5 soft, with 3 iterations. Remember, this only shows if you are zoomed in at 100% or more.

The next step is to turn the image black and white, because the colour is just noise. You can do that on the colour tab.

Finally, you can bring back some of the "pop" in the original image without raising the noise playing with Retinex and Haze removal.

Some of these steps are a bit computationally intensive, specially the last one, so if you have an old computer it may be a bit slow to find the right settings for you. A tip is to set the demosaicing algorithm in the settings to "fast" for lower zooms. This will speed up when you are tuning "global" settings, such as the white balance or the overall contrast, but will use the slower method when you are likely to see differences, ie, when you are zoomed in.

Detail: enter image description here

Whole (with a lot of JPEG compression, sorry): enter image description here

Settings used

  • Interesting workflow. Thanks for sharing! Feb 5, 2020 at 18:47
  • @EricDuminil RawTherapee is really powerful, and the documentation, RawPedia, is extensive. Plus, it is free.
    – Davidmh
    Feb 6, 2020 at 7:28

A lot of noise can be removed. I used unprocessing to denoise it, which is probably an overkill approach for this. denoised image

If additional sharpening was applied, maybe this image could be passable.

It is however evident that your picture is not entirely sharp to begin with, and as others have stated, going back and reshooting with a stable tripod and a lower iso would be the better choice.

  • It may have gotten rid of the sensor noise, but it introduced new color distortion: the upper-right corner of the image has a greenish cast, while the bottom-center is reddish. The actual Moon is a nearly-uniform grey.
    – Mark
    Feb 3, 2020 at 23:14
  • I'm afraid those colors are present in the original raw file. There are large amounts of red, also in the background which I think come from some ambient light. The reason my image looks slightly different, is because I reprocessed the raw file.
    – Atnas
    Feb 4, 2020 at 9:26
  • You might have achieved much better results if you converted the image to grayscale before denoising (or even demosaiced directly into grayscale, like with RawTherapee's "Mono" algorithm).
    – Ruslan
    Feb 4, 2020 at 19:57
  • Actually, even simply converting your result to grayscale gives me this. Seems not bad.
    – Ruslan
    Feb 4, 2020 at 20:28
  • @Ruslan actually the method denoises the picture before debayering, so this should not be a problem (although I think I misaligned the bayer pattern, because I did it quickly)
    – Atnas
    Feb 5, 2020 at 10:59

The ISO noise isn't the problem, it's shaking. There's definitely ISO noise but the shaking means that you can't clean it up. You can photoshop an image out of this pic but it won't be the real thing.

Here's the picture with saturation blown out. The banding is all bent and distorted from movement and the noise itself is turned into vertical lines. ISO noise bands are straight and the noise is dots not vertical lines.

enter image description here

You can't use the normal fixes for ISO noise for this because of the shaking. If you remove the color you're further blurring the picture. If you try the old blur and resharpen trick you'll just be losing the real edges and generating fake ones.

enter image description here

For comparison, here's an example of ISO noise with saturation blown out to 100-percent. See how the noise is dots and the banding is even lines?

enter image description here

If you put a little motion blur on that noise you can see how the noise starts to look like the noise in your pic. The even noise and even banding gets bent and vertical lines are made out of the spots:

enter image description here

  • Pretty sure as when I move the iso from High 2 to 1600 that clears up, that introduces motion blur from slow shutter speed thou.
    – lijat
    Feb 3, 2020 at 7:10
  • At 25000 IS0 you haven't got 8 bits of dynamic...
    – xenoid
    Feb 3, 2020 at 7:33
  • 2
    Ohhhhhh you're using an ISO setting. Just because it's an ISO setting doesn't mean it's creating ISO. Your picture is distorted way beyond ISO. That horizontal banding and breaking is not a color distortion, it's physical movement or something wrong with the camera. To shoot the above shot, there's no wind, the tripod is as short as possible, camera is on a timer and I had to get off the deck and stand still inside. The deck is on a roof, so it's solid, but if I was standing near the camera perfectly still you could see the shake in the pictures
    – moot
    Feb 3, 2020 at 17:18
  • @EricDuminil again, if this is caused by an ISO setting, that doesn't make it ISO. We have to reserve ISO for the specific type of distortion. The poster says there's shaking and motion blur... "I had to hold the camera, on a tripod, and time when I took the picture. This nessitated a fast shutterspeed to overcome camera shake at 3000mm"
    – moot
    Feb 4, 2020 at 23:45
  • @EricDuminil we're talking about ISO noise, sorry if me not typing the entire term out for you is confusing. If you can't see how the banding and noise is different in shot, then you can't. If you wan't to open a question on the difference, post your monkey pictures and I'll explain. They're so far from exactly the same I'm not sure your link is correct
    – moot
    Feb 5, 2020 at 17:38

Focal length is not everything.

You made a mistake by using 3000mm focal length. The truth is, 400mm is plenty. I took this picture of the moon with a 400mm lens:

picture of the moon with 400mm lens

If you use a longer focal length, you have poorer light collecting ability unless the lens is very huge. Every teleconverter reduces the resolution of the lens system. Also, you start to see the effects of atmospheric distortion.

I'd say what you are seeing are the effects of a poor lens system, not the effects of ISO noise.

  • What camera where you using and how big can you display that?
    – lijat
    Feb 6, 2020 at 19:46
  • @lijat Canon EOS RP, 26 megapixels, and this has 1:1 pixel mapping so you cannot magnify it any more.
    – juhist
    Feb 8, 2020 at 6:55
  • Then your 400mm lens is sharper than mine (soligor 400mm f6.3) or in the case of the 400mm lens it could me missing focus, hard to focus when the moon is that small in the viewfinder (liveview dont work so well in bad lighting on my camera). I have done more tries with the 1000mm mirror lens and 3x teleconverter and gotten better but not good results, a drop in outside temperatur might have helped as well as some of the non iso related tips. Next thing to try, a m43 body (gx9) with 20mpix sensor on the mirror lens. Higher pixel density might help. As well as no AA filter.
    – lijat
    Feb 8, 2020 at 7:08
  • 2
    @lijat My Canon 400mm f/5.6 is indeed a very sharp lens, and I used autofocus for this picture. It was shot at f/11. I think the lens is already sharp at f/5.6 but I used f/11 because it's easy to remember the f/11 rule.
    – juhist
    Feb 8, 2020 at 7:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.