Personal View site logo
GH2 In-Camera Noise Reduction Comparison
  • I performed an extreme example with ISO 3200 that shows the difference between the GH2 in-camera NR at -2 compared to 0.
    IMO I would always leave the NR at -2. The loss of detail (increased mud) and the blockiness is far worse than subtle fine noise.

    ISO 3200 66Mb/s AQ4
  • 24 Replies sorted by
  • That is very convincing. Can you state the file sizes, or bit-rates, assuming both videos were shot identically except for the NR setting.
  • Can you try it with 44Mbps AQ4?
  • Thanks for confirming this. I've also noticed that having the in camera NR set to zero results in a strange inconsistent red heavy flickering noise pattern which is pretty much impossible to remove. I quite like it (but I'm weird) and played with it a bit recently, I used Nukes deflicker node to deal with it as any kind of denoising isn't going to do much with it at all. The flicker is more obvious in the original file if you download it.


  • @aljimenez
    The camera was never moved. The only thing different is NR at -2 verses 0.
    Everything else was a constant, lighting, ISO, f/stop, shutter speed,....

    Also,
    I realize this is an "extreme" example, but I have seen this in shadows where the in-camera NR just destroys image detail.

    Although I do not have enough tests to confirm, it does not appear that the hack is helping much for the in-camera NR. But, the hack certainly helps post NR solutions such as Neat Video! The increased image quality now displays fine detail such as noise which make noise removal a lot cleaner process!

    I'm just pointing out for those who are looking for the cleanest image possible, leave NR at -2 and handle your NR in post.
  • Could you try the same test with Sharpness and Contrast? I would like to see how those affect the specific test you are doing. I would expect all of them to negatively affect the image except maybe contrast. I still have not totally figured out how that setting works.
  • Makes sence. The GH2 manual does warn about loss of detail with NR. It makes sense to turn off/reduce in camera processing if you can when hacking; the less unnecessary processing the camera has to do, the more stable it should be/more CPU cycles it has to handle high bitrates, etc. It can make a difference; settings that the GH2 is very stable with with a legacy lens can be very unstable when you use a M4/3 lens which requires distortion/aberration correction, for example.
  • @Ptchaw You may have missed a post by @cbrandin about how a bit of on camera noise reduction will in fact yield better quality in the end. If I recall correctly his argument, some noise correction in camera will free bits that normally go to compress noise accurately to be used for more accurate compression of what is important in the image. His idea is that if you are going to remove the noise in post, the camera wasted bits compressing the noise. Any noise comparisons really should be done after doing post on the image to make decisions... Al
  • @aljimenez Interesting, I missed that. Did he actually manage to confirm that with tests? Though that might only yield better results if you were planning to do NR in post, rather than no NR. Otherwise the image above would suggest a loss in detail.
  • Chris is right assuming the end results using the in-camera NR are good enough.
    The problem i'm seeing is the current GH2 in-camera NR will never be good as post solutions.

    Also, it would depend on where in the signal chain their NR takes place.

    One last thing, all noise reduction programs cause "some" loss of detail. Some programs are just better than others. The in-camera soultion has no control of what it thinks is noise. It's a one button knob. Programs such as Neat Video have a lot of control with many parameters.
  • I saw @cbrandin say that too but my own experience with in camera NR is the same as @proaudio4.

    After chris said that I ran a lot of tests with in camera NR and the results were all pretty awful. In fact if you're shooting a scene where the light changes at all, even at a higher bitrate, it can do all sorts of whacky things. I think we all agree that at a higher bitrate with NR set to -2 the noise is a lot less smeared, and therefore can be cleaned with your denoiser of choice with a lot less artifacts and with much better end results. Another thing I noticed is that it didn't actually free up any bits, as the theory goes, so as to not waste bandwidth compressing noise, but can use it instead to compress the actual image detail. In fact the average bitrate dropped quite a bit on the same shot, therefore there is no benefit. What is happening is the detail is being removed by the in camera NR, so it ends up compressing less, as the detail has been lost.

    Edit : @proaudio4 I think vitaliy and chris have both said that its at the front of the chain, before any compression. So from that, and what we've seen, we can just assume its not a very good NR method thats being employed.

  • +1
    Stray.
    You are absolutely on the money with this one!
  • There haven't been any honest tests when it comes to NR. NR always removes detail and is usually followed by sharpening. If you do no NR in the camera then you busy the codec with reproducing noise rather than reproducing aspects of images that may be more beneficial. I would like to see tests where a little NR is done in the camera, followed by additional NR done in post, followed by sharpening is compared to no NR done in camera, followed by all all NR being done in post, followed by whatever sharpening is necessary to compensate for it. I suspect that doing a little NR (and I mean just a little) in camera may yield the best results. We shouldn't be comparing doing all NR in camera with doing none.

    Chris
  • @cbrandin I completely agree, it is an area that needs a lot more testing. Also, yeah, the noise is obviously going to busy the codec more. But I feel its going to be down to each persons individual workflows, their aesthetic needs, and each shot, judging from the NR and noise removal experiments that I've done. Because of that I feel we're not going to reach any kind of consensus other than 'its something to consider when shooting at higher ISOs'. So its something for each of us to test I feel, and to come up with something that we're happy with individually. As I discovered with my test above (one of many I've done with in camera NR and various noise removal methods in post) I found something with in camera NR that I liked, which I'm pretty sure noone else here would, or would tolerate (the denoising flicker). Also that kind of thing is not something that can be removed easily with just denoising in post. I also found with that shot that shooting with NR set to -1 was pretty ugly, subjectively to me the shot with it set to zero was the best and the one I felt like working with. Shooting at +1 or +2 it looked terrible to me.

    My original guess about the in camera NR filter was that it was a cheap and cheerful linear smoothing filter, however that wouldn't to my mind create the flicker we're seeing (unless it is a 2 pass process with linear smoothing being in the second pass). What would would be some kind of median filter where its getting the intensities wrong somehow under certain conditions. The fact that the flicker always has more intensity in the red channel, and that, as said in other threads, this camera is more noisy in the red channel is a worry. Surely Panasonic would know that its at it's worse in the red and would weight their NR algorithim accordingly. Or maybe they are, and that, coupled with higher bitrates, is causing the odd flicker on certain shots when the NR throws incorrect intensity values. Also its hard to know if the stress of runing the camera at higher bitrates is itself effecting the abilities of the in camera NR (though it probably isn't, I'm guessing).

    I'm lucky when it comes to denoising in post as I have a few options, from PFCleans approach, to Neat Video and finally Nukes new wavelet based approach. Each produces different end results under different conditions, and the choice of which, and the solution I end up being happy with is subjectively mine. I don't think myself that any in camera NR is going to be acceptable to those who do NR in post, as what it does is to create something thats more difficult for any post NR tool to deal with. I said 'I don't think' at the beginning of the sentence, I do agree that more tests need to be done here. Thing is though NR is full of vagaries. I've never found that I've got good results running footage thru mutiple NR solutions in post (unless one of the NR approaches I've used in the chain is used in a really, really foccused way; as in linked to a single colour channel, or to a single band of noise frequency), it usually just creates problems further down the NR chain. So the idea of doing a little in camera followed by doing more in post fills me full of dread as to the problems people are going to get doing that, these will be a serious of image grabs and Neat Video settings I don't think I'm going to want to see ;) .Its just going to confuse more than reach a conclusion I reckon. Stick to what @proaudio4 has done at the start of the thread, strict in camera NR comparisons at different ISOs and bitrates, thats information we can use and react to usefully.
  • Sorry, feel I need to add my own take on NR after doing it in post for too long. It is always shot specific as I said, and so the solutions are too. Also, with some shots you just won't be able to remove noise entirely, your goal is to get it down to a level where it ceases to be distracting.I think there is too much focus being given here to just popping a denoising filter onto your footage when there are other (and much faster) ways. Firstly you can mask out areas of your shot either by colour ranges or channels, or by roto shapes, and then have seperate denoising settings for each area and then reconstruct the image. Also there are other options you can use on these different areas, channels and ranges, such as diffusion, exposure, clamping, blurring, convolutions, colour correction etc, etc lots of things work other than or as well as using a denoising filter. Also, if you have a film degraining tool that can give better results with different shots.
  • If been shooting various tests at 66mbps going through the whole range of NR at 1000 and 2000 iso... and found generally that -2 produced a more even tone of grain... as opposed to the other NR settings which tended to cause the noise to dance around the screen drawing more attention to itself... but who knows under some circumstances +2 NR might look great, its nice to be able to test the location and lighting if you have the time, but in a push I'd prob leave on -2 and remove noise with software.
  • @proaudio4 what was your contrast set to with the test above ? I'm noticing, while testing in camera NR, that bringing contrast up improves the cameras NR. I get the denoise flicker with NR at zero and Contrast at -2, and it seems to vanish with contrast also set at zero. This also seems to be the case with high contrast scenes in general.

    Edit : All said though, it still looks horrible compared to NR set at -2.
  • Stray,
    I'm fairly sure this test was at 0 for both Contrast and Saturation.
    Interesting find on your end.
    I'll have to do more testing with contrast at 0 verses -2.
  • I don't think there is anything inherently good or bad about any particular contrast setting. It's totally dependent on the contrast of the subject. Assuming you'll be doing some post work, I contend that the right contrast setting is whatever setting fills the histogram. With low contrast subjects a higher contrast setting is better, with low contrast subjects a lower contrast setting is better. For maximum quality you always want to use the full dynamic range available without clipping. Same thing with saturation - although you can't use the histogram to see that.

    Chris
  • @cbrandin Yep agreed, what I'm saying is it seems from what I've seen that the contrast in the scene, and the contrast setting in camera has a strong effect on how the in cameras NR performs. If you do set contrast badly, as you say clipping the dynamic range, there is more chance of flicker and the results out of the in camera NR get a lot worse. This isn't a surprise really, and I've no idea why I didn't test it before (probably because I'm trying to tweak with the camera at the same time as working on a project, can't help myself, it's just too interesting), as NR post routines exhibit similar behaviour too.
  • @Stray

    Sure, if you think about it: if the contrast setting is too low you are essentially increasing quantization (a bad thing) by not using the entire available dynamic range.
  • Are we are talking about low lighting case?

    Normally I don't wanna jack up the ISO too high. I'd rather get underexposed clips. When I boost midtone gray during PP, ugly noise become more visible.

    Saturation -2. It's underexposed in the first place. I wouldn't get great color. Just suppress it to keep the color noise down.

    Contrast -2. I wanna keep the midtone gray details more in this case. Crush the blacks during PP.

    Sharpness -2. It's not properly exposed. No need to depend on in-camera sharpening in this case.

    NR -2 or -1. Low lighting again. Less details to render. Possibly more noise reproduction might help in this case.


  • @stonebat

    Your points about underexposed or under-lit cases are well taken. I was, of course, assuming relatively normal lighting conditions.
  • Hello All,

    This manual is better than nothing, but it doesn't have everything in the contents or index so finding information is difficult (not intuitive).

    I see a setting for "Long SHTR NR" but that doesn't function while shooting movie clips, so that's not a factor.

    Where do you find the settings for the internal NR?

    Thanks!

  • Very helpful, thanks for posting.