5

This question is motivated by the fact that I'm trying to decide between the the Canon 5D Mark III and the Nikon D800. I really would prefer the one that will perform better in low light.

As far as I can tell, it seems that various ISO comparisons show that the Canon does better. My understanding from reading questions like this and articles like this is that it is easy to come to the conclusion that "Oh well Canon will definitely do a better job because the sensor is basically the same size as the Nikon and it has way fewer megapixels". But I'm worried that that argument is flawed.

It should be noted that what I really care about is which will have ultimately better results after post processing (and I will probably only use Lightroom unless you can give me another simple technique for reducing noise).

Are there other factors at play here where the Nikon will ultimately perform better in low light? For example, are there modes where you can reduce the pixel count in the interest of getting less noise? Does the higher pixel count make it easier to reduce noise in post? Comparing image quality while varying ISO seems like it might be like comparing apples and oranges, so I'm hoping that someone here can teach me what I really need to look for and how it applies to these cameras.

|improve this question
  • More context: I currently use a Nikon D7000 with a 24-70 f/2.8 lens. I get pretty great shots up to ISO 6400 in pretty low light when I'm wide open at 2.8. I like low light stuff and I even find this limiting because there are times I wish I could use higher shutter speeds. I was already planning on investing in better glass (f/1.4 stuff) but I also wanted to upgrade to full frame so I am also considering camera performance. Honestly, if Canon does better and also offers f/1.2 glass AND has smaller images (who needs >30MP?)... I think I might be willing to sell my stuff and make the switch. – Tom May 16 '12 at 3:32
  • Do you care about noise as seen in prints? Or do you only care about per-pixel noise? (If the latter, why?) – James Youngman May 16 '12 at 12:29
  • 2
    The difference between f/1.2 and f/1.4 is only half a stop. I find that the circumstances in which I even want to use f/1.4 or f/1.8 are very very limited - I have primes in that range, but the DOF is too small even for most portraits at those apertures, at least for the portraits I make. – James Youngman May 16 '12 at 12:30
  • This is a hugely divisive, and subjective question and I personally think you won't get any meaningful answer from it. Just a bunch of people squabbling over the old Canon/Nikon "who is better", "my dad's bigger than your dad" debate. I've had the luck to use both the Canon 5DmkIII and Nikon D800 and both are incredibly good, very capable cameras. If you're already invested in glass in one system, go with that. Focus not on "which is better" and just get out there and take the photos. – Mike May 16 '12 at 13:44
  • you can look at this question for noise reduction software photo.stackexchange.com/questions/483/… – K'' May 16 '12 at 14:24
8

All other things being equal, bigger pixels give a better performance. In the real world though, all other things are extremely rarely equal.

However, given both are latest full-frame models, one would guess the 5D Mark III does better (which I just reviewed it here and you can see full-resolution samples at all ISO sensitivities) in terms of high ISO. Keep in mind that both Canon and Nikon have higher end models even more optimized for low light, the 1D X and D4.

Now, as you reduce image resolution by downsampling, you can effectively simulate bigger pixels. The site DxOMark which measures sensor performance works by comparing non-resolution limited output normalized to 12 megapixels. Under that measure, their numbers for the D800 at high-ISO are better than those of the 5D Mark III. At full-resolution though, results are mostly likely reversed (which I expect to find out but the D800 has not arrived yet).

|improve this answer
  • what is downsampling? – K'' May 16 '12 at 14:26
  • @AkramMellice down-sampling is a fancy way of saying resizing smaller. – Ian Lelsie May 16 '12 at 15:24
  • 3
    Bigger pixels give better performance on a per pixel basis, which is next to meaningless. It might make you feel better when you view your images at 100%, but in real world use cases comparing images at the same final output size makes a whole lot more sense. Total image noise matters far more than per pixel noise. And all things being equal smaller pixels give better performance with this metric as you capture more information from the same amount of light. – Matt Grum May 16 '12 at 17:03
  • @MattGrum: Totally agree with you in general about normalizing. Only caveat, I would say, would be when normalizing image size is detrimental to a photographers goals. To clarify, if you buy a big megapixel camera, like the D800, its most likely unlikely that you intend to use such a camera to downscale to 900 pixels long-side for web viewing. It IS likely, however, that you are buying such a camera for critical print work, proof work, etc. In which case, every last original, unscaled pixel is important. Per-pixel noise is important then, is it not? – jrista♦ May 16 '12 at 18:27
  • 1
    @jrista ... If you normalize by upscaling then the situation is different as the low MP (and lower per pixel noise) camera will yield an upscaled image with a lower intensity but more blotchy noise, and more importantly less detail. But the key point is you can always downsample your high MP image if per pixel noise is suddenly important for any reason, so there's no disadvantage when it comes to noise with more smaller pixels (unless the per pixel noise is much higher, which it isn't with the D800). – Matt Grum May 16 '12 at 22:29
5

The correct way to determine which camera is better in low light (with respect to image noise) is to compare images at the same final output size. This is the only comparison that makes sense, and is what DXO mark do when the normalize for print size.

I [hypothetically] go on holiday with a Nikon D800, and a ten megapixel Nikon J1 for when I don't want the hassle of a big camera. Now when I go to print my images or display them on the web am I going to have the D800 double the size of the J1 images? Nope. I shot images from both cameras for the same purpose and will therefore reproduce them at the same size.

Now large pixels may capture more light, and give you less noise per pixel, however this is a pretty meaningless metric. Per image noise is what matters, and when you take an image with small pixels and print or display on the web at the same size you should get a similar amount of noise.

In fact for identical technology it's possible to get better performance out of the camera with smaller pixels. Imagine you half the pixel size. Now you can always add up the light collected by four smaller pixels to replicate the effect of one large pixel, so there's no loss attributed to the smaller pixels, however with smaller pixels you are capturing more information about not only the total amount of light, but where that light is falling. This extra information can be exploited by noise reduction algorithms to improve image quality.

Now things are never quite equal in the real world and read noise and microlens performance differ based on pixel size. However, referring to your question, the Canon 5D mkIII and Nikon D800 are close enough in low light noise performance, when taking the whole image into account, that I wouldn't base any decision on this factor.

tl;dr The differences in resolution, price, lens selection, ergonomics, etc. are all far greater than differences in low light performance so I would look to these instead.

|improve this answer
  • 1
    As you know, some noise components scale differently than others when using more or less pixels. eg there may be noise per channel (or pixel) which is related more to doing the task (say ADC conversion) than to the pixel area. Worse the noise may be about constant per channel so eg 4 times as many pixels = 4N pixels = 4 X the noise, but the signal to noise per channel will decrease with th pixel increase so that a degradation of between 4x ad 16x will result from a pixel size shrink of 4X for this noise component. Other components may be linearly affected or not affected at all overall. – Russell McMahon May 16 '12 at 18:25
  • 1
    @Russel sure the real situation is much more complicated than the purely theoretical case when you take read noise into account as it also depends on the sensor architecture (the Sony chip in the d800 is far better in this regard), the important point is that the cameras in question were actually closer in noise than they are in other areas. – Matt Grum May 16 '12 at 22:16
  • @MattGrum - Thanks for the good explanation. To make sure I understand: it sounds like more pixels = more noise but also more information. The extra information actually makes it easier to remove noise later in post processing? Do you know if noise reduction algorithms already leverage this (you seemed to speculate they could leverage the info :-)? Would you say that if I took pictures with the mark 3 and d800 that looked very similarly noisy when I import them into lightroom that I'd be able to clean up the noise better with the image from the d800? – Tom May 18 '12 at 6:10
  • I guess another important thing that came out of all these answers is that there's no in-camera solution this problem. For example, you can't flip a switch and make the sensor optimize for low light. Also: I guess using "crop mode" is much worse because you actually don't use the entire sensor. – Tom May 18 '12 at 6:28
3
  • Context: I do not own a Nikon or Canon DSLR. I have had a long string of Minolta & Sony cameras. If Sony do not make their cameras make the sensors that they sell to Nikon work as well as Nikon does at high ISO, and soon, I'll be changing to Nikon or Canon. I value good quality high ISO capability and the ability to take photos at a rate that the situation requires.

See the DXO Mark comparison between the D800 and 5DMkIII that Itai referenced.

Also see the DXO Mark high ISO performance comparison chart for all the cameras that they have tested to date.

The vertical (Y) axis is the maximum ISO at which a specified set of quality measures are met. This is ~~ the max ISO that you can take a photo in normal conditions and expect a "reasonable quality" result. Their sit explains the above in more detail.

As Itai notes, DxO "normalise" their results by scaling to a 12 megapixel image. They scale the signal to noise results down by a factor of square_root(sensor mp / 12 mp). This effectivelu scales the ISO result up by the same amount. So eg a 24 mp A700 would have an increase in the catual rating by a factor of (24/12)^0.5 = sqrt(2) ~= 1.4.
And a 36 mp D800 would have its ISO result "improved" by sqrt(36/12) = sqrt(3) = ~ 1.7.

DxO ISO results for the top 35mm cameras and their megapixel count are as per the list below. IF you believe hat DXO's scaling method provides a good match to reality then you can use the chart below for comparison. This is derived from DxO's table here (same ref as above) and camera mp sizes. While the D800 is shown as ahead of the 5DMkIII by 2853/2293 ~= 1.24:1 that's about 0.3 of a stop. Notice it, even by pixel peeping, try as you may, won't you. The now venerable D700 sits happily between the two. Remember that his is when the image has ben adjusted to 12 mp by whatever magic - so the awesome 36 mp of the D800, if you had set everything jus so o achieve it, has been obliterated. The 4 year old D700 is still right amongst them. The D3s is just noticeably above the D800.

The second chart has had the ISOs adjusted back to what is actually measured before allowance is made for sensor size. I divided the reported ISO ratings by sqrt(12/mp).
Results are sorted in order of raw ISO.
D3s is top. Again. No surprise.
D4 is second. You'd hope so.
D700 & D3 (same sensor in fact) trail just noticeably behind.
THEN
A small cliff before you get to the 5DMkIII, D800 (gap so small as to not be noticeable) , D3X, 5FMkII and a slow ail down to the D5100 & D7000. The should be ashamed of itself Alpha 900 shares bottom place with the best of the APSCs.

While their is a mini cliff between per pixel D3/D700 amd D800/5DMkIII of about 2300/1680 =~ 1.4:1 this is still only about half a stop!.

Conclusion: If you want good low ISO resuls and you believe the tests and the method of scaling gives a realistic basis for comparison then there is only half a stop across the range D3 - 5DMkIII - D700 - D800 - D4 - D3s in that order. D3s is still king.

If my eye-brain system is to be believed (and why should it be ;-) ) there is somewhat more than that between them.

Re Tom's D7000: It is about 1 stop down on adjusted ISO wrt D800 or 5DMkIII and somewhat more than half a stop on raw ISO. Not vast in either case.

I'd strongly suspect you'd see a greater difference that that in real life - based on my peering at comparative high ISO images for other cameras.

However - if you want real world performance and are interested in buffer size and peak shooting rate I'd look very closely at comparison tests in environments of relevance to you. I've seen comparisons of the two cameras in situations where the nominally superior Nikon performed very badly indeed in tests that were relevant to me - not on noise, where it was comparable to the Canon, but in usability to take actual shots in real world situations -

frame rate in short bursts
effects of buffer size and file size and data transfer rate, and
motion tracking.

In at least the first two the D800 is behind.
As much as I hoped it was going to be a D700 successor, it's not.
Vanishingly few people would have chosen the D700 for studio shots.
When comparing it to the D800 or 5DMkIII it's in extreme lower light situations and/or where there's plenty of action needing bursts of rapid shooting and possibly as close to semicontinuous shooting on occasion.

My personal conclusion is that the D700 is (still) looking good (4 years on), a D3s would be very nice, and I can hope that Canon has got its act together with the yet to be judged 1DX.

enter image description here

|improve this answer
2

Disclaimer: I own an 5D Mk3 and I shot 10000s photos with it already.

Itai is right in what he says. But the answer isn't limited to sensor's performance.

For me better in low light means also a superior AF engine which is capable to focus accurately in low-light.

Also, I suppose that better in low light means that you're looking for an event, photo journalism camera and not for a studio one. And here also comes to play other things like the number of frames per second, handling / ergonomics, the LCD screen performance in low light, things on which Canon is superior.

But if you're for taking landscape shoots with extreme DR in low-light, then perhaps Nikon is better. Except if you use Canon's built in HDR mode on a tripod. (If you are serious about landscape you need a tripod anyway).

I also thought between the two and I reached at the conclusion that Canon is the best journalism, event, wedding-like solution whereas Nikon is more geared towards studio.

See what your needs are and choose.

|improve this answer
  • 4
    whether Canon is superior or not when it comes to ergonomics is a highly personal opinion. For me Nikon beats Canon hands down there, enough so I would be more than willing to put up with somewhat (and the difference is likely marginal) lower technical performance if any. – jwenting May 16 '12 at 11:56
  • I think both of you are basically saying the same thing even though due to the subjective nature of the considerations you have a different conclusion. The point is that technical measurements of noise in low-light situations is only one aspect of the whole picture. – mattdm May 16 '12 at 13:44
  • @user, right. What is "best" depends on what the criteria are. For me, being able to AF counts a lot, especially with the evil kit lens that are slow at AF even in a fair amount light. – Pat Farrell May 17 '12 at 4:47
  • Good point about autofocus. How much of that performance is based on the camera vs. the lens? – Tom May 18 '12 at 6:13
  • 1
    @Tom: Both, but more is to the camera: Camera has an AF engine which has as main characteristics: 1.) The total number of AF points (61 - Canon, 51 - Nikon) 2.) The number of higher accuracy Cross Type points (41 - Canon, 15(!) - Nikon) 3.) The number of low-light extreme performance Dual Cross Type AF Points (5 - Canon, 0 (zero!) - Nikon). This is what camera gives. But what lens can use is based on several factors, from which the most important is, by far, the lens' aperture. As a rule of thumb, the lenses with a maximum aperture of f/2.8 (or faster) can use the entire AF engine. HTH – John Thomas May 18 '12 at 6:35

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.