Empirical observation #1: Cameras with higher resolutions, for a given sensor size, are harder to hand-hold and get a good result than those with a lower resolution.
And I define a good result as one which the image is critically sharp at 100% actual-pixels view, with no motion blurring or other artefacts such as slightly missed focus (we’ll get to this later). Motion blur is when the camera moves by a greater amount than a pixel during the time when the shutter is open; if it’s less than this, then you see either hard edges on your subject as a perfect sharp edge, or with slightly lower contrast than expected due to the image projected by the lens on the sensor plane ‘spilling over’ into the next adjacent pixel slightly when the camera moves.
The simple reason why higher resolution cameras are less forgiving is that for a given angle of view, the camera with the high resolution will put the greater number of pixels per degree of field of view on the subject. This means, that less angular movement is needed before it becomes visible as blur on the image – i.e. you need to be more static than with a lower resolution camera to avoid seeing the effects of motion blur at the pixel level.
But what about the image when viewed as a whole at the same reproduction size? If we started out with a huge number of pixels with both – say 24MP and 36MP and resized both them down to say, 4MP or so – we would not see any difference between two cameras assuming one assuming both were equally downsized. This is because the downsampling process has enough information for each pixel to accurately determine its color and luminance value and therefore appears as true detail. If we downsize 36MP to 24MP, it might or might not look better: a lot of the outcome depends on the interpolation process. It isn’t a ‘clean’ situation where each new, downsized pixel has multiple pieces of information (the original pixels) to form an average that would be the same whether you used 8 or 12 pixels originally. Since 36 doesn’t divide cleanly into 24 – you get 1.5 old pixels per new one – there’s always going to be some guesswork as to precisely how that half pixel is allocated. And depending on the algorithm, any one of the following might happen – blur edges; stairstep artefacts; haloes or abrupt transitions; odd discontinuities in diagonal lines.
Let’s take an example: if you have a scene shot with 12 (8.45um pixel pitch) and 36 MP (4.88um pixel pitch) cameras, and both are equally motion-blurred. The degree of camera shake may be less than a pixel on the 12MP camera – let’s say 5um for argument’s sake, which will still produce a critically sharp image. However, the 36MP camera is now blurred by more than one pixel, which definitely will not look critically sharp at 100%. Yes, we can downsize – but you are downsizing not to 1/2 of the original linear dimension (which would yield 8MP, and bin the two blurred pixels into one sharp one that looks perceptually okay), but to 1.7x of it – and once again, we run into potential issues in trying to fit discrete, nonlinear data into a linear continuum.
This is one of the reasons why less pixels might actually produce a perceptually sharper/ crisper image for a given reproduction size, providing that this size is reasonable for the amount of resolution you’ve got in the smaller image.
The upshot of all of this in practical shooting is that you’ll need to use a higher shutter speed than you expect, which lowers your shooting envelope by a corresponding number of stops as noise increases as you ramp up the sensitivity to keep shutter speeds high. Perceptually, if you’re going to be shooting under already borderline conditions to get a critically sharp image with a 12MP camera, you’re not going to get one with a 36MP one – even if you print them at the same size.
Empirical observation #2: Cameras with higher resolution require more precise autofocus.
Comparatively, this one is very simple to explain: if the camera misses focus, the resolving power of the lens is diminished. You need to have the ability to resolve slightly smaller structures than the pixel pitch of the camera in order to create a perfectly sharp image at the sensor plane. If you miss focus by an arbitrary amount – let’s say 1% of your subject to camera distance – this may result in a loss of resolution corresponding to the inability to resolve structures below 5um in size. For our 36MP, 4.88um pitch camera, this will appear to be a blurry shot. For our 12MP, 8.45um one, things still look critically sharp.
A lot of what we’re seeing isn’t necessarily worse autofocus; what’s happening is that we are now visibly reaching the limits of tolerances and precision since sensor quality is now high enough to resolve at or close to the previously acceptable tolerances.
A lot of the reasons why an image appears sharp or not has to do with the reproduction medium, as well as how our eyes interpret information: an image with hard edge is seen to be ‘sharper’ than an image without one, even if there is less fine detail or edge definition. The fact that digital capture splits out our images into distinct zones – the pixels – doesn’t help this at all, either. It’s important to remember that a slightly blurred image (depending on the degree of blur, of course) from a higher resolution camera may technically still have more detail, our eyes don’t necessarily pick up on this; it’s a perceptual thing. Bottom line: don’t use more pixels than you are capable of handling properly for a given situation. It’s one of the reasons why I don’t shoot the D800E for casual work…but it’s also the reason it’s my camera of choice when I have full control of the execution. MT
____________
Visit our Teaching Store to up your photographic game – including Photoshop Workflow DVDs and customized Email School of Photography; or go mobile with the Photography Compendium for iPad. You can also get your gear from B&H and Amazon. Prices are the same as normal, however a small portion of your purchase value is referred back to me. Thanks!
Don’t forget to like us on Facebook and join the reader Flickr group!
Images and content copyright Ming Thein | mingthein.com 2012 onwards. All rights reserved
Ming Thein | Photographer 
I think you’re missing an important point here, one that many fail to understand. If you look at a D800 image at 100% on a typical screen (say 24″), you are effectively blowing that image up to about 6.7 feet wide and 4.5 feet high, and you’re looking at it with your nose 12 inches from the screen. Not a very realistic situation. Or, to put it another way: in the good old days of “wet” photography, we checked the sharpness of a slide on a lightbox with a 10x loupe. Viewing a D800 image at 100% is equivalent to using a loupe with a magnification of about 60x. No slide from any camera/lens/film combination would ever have appeared sharp at that magnification.
As the pixel count increases, the 100% view means we are enlarging our images more and more, to the point where it becomes meaningless. This is why images from high-MP cameras seemingly appear less sharp than their lesser brethren. It’s because we’re blowing them up more and more, to the point where everything appears blurred. This will not change until someone invents the fractal digital camera!
It depends on your output method. I am resolution limited in print by 16×20″, and yes, I can see the difference quite clearly.
Not sure what it is you can see the difference in, here. But here’s another thought on why your argument is not valid. You talk about motion blur in relation to a pixel size of 5μ. You should know that this is the size of the airy disk, or circle of confusion, for a lens at f4. This is the diffraction limit imposed by the lens, and has nothing to do with the sensor. At f16, the airy disk is about 21μ, covering at least 4 of these 5μ pixels, though the sensor’s anti-aliasing filter will mitigate this to some extent. Diffraction will, however, have a visual impact before reaching this diameter.
It is not motion blur which is limiting the sharpness in your example, it is diffraction limiting, and that is why a 12MP sensor may appear sharper than a 36MP sensor, when viewed at 100%. The 12MP sensor photosites are physically larger, and therefore diffraction limiting is less obvious. This is simple physics. And it’s a very good reason not to pixel peep at 100%! All you’re doing is revealing the fundamental limitations of your lens.
There are a lot more factors at play when comparing sensors at different resolutions – different resolutions at the same sensor size will generally mean different generations of photosite architecture, for starters. Secondly, nowhere do I mention shooting at f16. And thirdly, I can see the difference in print – so can everybody else I’ve shown the prints to. Tell you what, we’ll each just do whatever we believe is best.
I use two Nikon cameras for the majority of my photography, a D700 and a D7000. I am curious to know how to apply your advice for downsizing to a divisible whole number of the native resolution, for web posting for instance, since that seems to be what you are advising (at least I think that’s what your suggesting). I believe the longest side of a RAW image from my D7000 is 4928 pixels. If I want to downsize to web post size anywhere between 800 and 1200 pixels for the longest dimension, what would that number be? For the D700 the native resolution longest side is 4256 pixels. So what would be the optimum longest side dimension for a downsize of a D700 image between 800 and 1200 pixels?
I just do a 25% downsample.
Well this confirms my findings as well…my motion blur has consistently increased as I have stepped up from M8 to M9 to now the M240 (the latter being the worse for me – maybe 24Mb is my limit!). I immediately sensed the difference on the first day of shooting but could not explain it until now….makes perfect sense…this is also the reasons I have decided to go with the Sony a7 as opposed to the a7r despite the bigger resolution…Thx for clearing it up
Two “empirical” observations and zero examples / data, not a good start. Also the argument about downsizing 36MP to 24MP being an unclean “1.5x” reduction is nonsense. Firstly what do you think the printer does when printing your images? The get resampled to a fairly arbitrary resolution internal representation. Same goes if you view your images in their entirety on a standard screen. Secondly if you don’t like this sort of “unclean” resampling that generates “blur edges; stairstep artefacts; haloes or abrupt transitions; odd discontinuities in diagonal lines” then you should ditch the Nikon D800/D600 and shoot only Sigma DSLRs with Foveon sensors (or better the Lieca M Monochrom) as Bayer demosaicing algorithms do an awful lot more tricky unclean things than a bicubic resample!
Thank you for the reply Ming and for thinking along Nick. So you think the problem lies in the focussing screen.. is there anything I can do about that to improve it so I have better chances of using my d800e as a handheld MF type camera? So far my search for a propper substitue for my MF film camera as a handheld documentary device has been a bit of a letdown (not quality wise, the d800e is awesome! But usability wise) Right now I am trying the Sigma dp2m as well as a MF substitute, and so far it seems to be a better option than the Nikon (the lens/sensor combo is amazing and it has AF while still being a discreet little camera.)
But still, any suggestion on how te get the most out of my d800e as a handheld device would be most welcome!
Cheers,
Ricky
An idea I’m going to try myself is to get one of those viewfinder loupes and use LV for MF work – at least that lets you brace the camera against your face for stability…could work. I too want to be able to use the damn thing handheld with the MF glass I’ve got!
Hi Ricky,
Keep in mind that even manual focus on a focusing screen is fraught with camera tolerances and inaccuracies.
The most accurate form of manual focus is to use a viewfinder loupe on the Live View display like Ming is describing above.
Kinotehnik, Zacuto, Zeikos and others offer loupes for that purpose.
The only accurate form of AF is CDAF performed by the image sensor.
Ming, thank you for this great article.
I have bought the d800e because I wanted to have an (sort of) affordable alternative to my Pentax 67 medium format film camera, which I use for my documentary work. I use the Pentax mostly handheld because it keeps me in a flow of things. And although some prints could have been a little sharper when I had used my tripod, in 95% of the cases it is just fine. The tonality and the overall feel of medium format film is still great. However, since I use the d800e, I find it so hard to use manual focus handheld shooting. The 95% I managed with my Pentax, maybe now just 20% is left when I use the d800e. After reading you conclusion: (Bottom line: don’t use more pixels than you are capable of handling properly for a given situation. It’s one of the reasons why I don’t shoot the D800E for casual work…but it’s also the reason it’s my camera of choice when I have full control of the execution.)
I come to think I might have used my d800e wrongly. Instead of treating it as a MF camera, I should be treating it as a technical camera.
What are you thoughts and experiences Ming? Any suggestions on how I still could use the d800e as a handheld manual focus camera? (by the way, I don’t shoot moving subjects, or sometimes very slowly moving subjects so I am not a street shooter)
I think your problem is the focusing screen. Modern DSLRs are optimized for brightness, not ‘snap’ – as a result, MF is very tricky. I use LV on a tripod, and yes, shoot it like a technical camera – it’s absolutely superb for that. MF image quality with the convenience of 35mm. For handheld work…I use an OM-D 🙂
Ming, would it not be possible for Ricky to do quality, hand-held MF work with the D800E if he followed your suggested guidelines of good hand-holding technique, use of high-quality lenses, making as sure as possible that the subject is in focus, and shooting at least 2-3 times the focal length? Wouldn’t that not improve his percentage of keepers?
It would, but that limits your shooting window dramatically.
That said, I’ve got a H4D-40 on loan and I’m doing exactly the same thing. 😛
Mathematics and camera shake.
While Ming is shaking a Hasselblad H4D-40 here is a theoretical comparison of how sensitive the Olympus OM-D, the Hasselblad H4D-40 and the Nikon D800 should be to camera shake. The calculations assume that you are framing a one meter wide subject. Key variables are sensor size and resolution. The focal length of the lens used does not enter into this calculation. The calculation determines how much each camera has to be moved sideways to shift the image exactly one pixel on the image sensor.
Here are the answers:
Olympus OM-D: Allowable shake = 0.217 mm
Hasselblad H4D-40: Allowable shake = 0.140 mm
Nikon D800: Allowable shake = 0.136 mm
Perfect: so the OM-D will permit more shake before you see the 1-pixel displacement. The D800e and H4D-40 are about the same, which is what I’m finding now. Mirror lockup on both helps a LOT: the D800E has some fine vibrations, the H4D-40 kicks like a mule.
Hi Ming,
It would really be helpful to see what a 1 pixel displacement looks like in an image. Also, are you seeing this displacement at 100% magnification or some other value. I’m not as technically proficient as many of you are so it would help to have an example. Thank you.
I’ll see what I can do. The problem is that this test is not really scientifically repeatable – I can’t shake my body precisely one pixel 🙂
Nick,
I write one pixel, but I meant one promille. Like Ming said, accurately moving the camera one pixel is not easy to do. The calculations above are based on producing a one promille shift of the image, in relation to the total frame. Compare that to shaking or moving a one meter wide target by one millimeter.
“Clearing up the myth of higher resolution, shot discipline and image quality once and for all.”
To compare images out of a 16MP mirrorless camera (with a highly accurate image sensor CDAF) to that of a 36 MP DSLR (with secondary image plane PDAF and known autofocus problems) is not a good basis for any kind of a conclusion.
We all agree that the Nikon D800(E) comes in several flavors, including with a light, medium or heavy PD autofocus problem.
Some Nikon D800 shooters report excellent results handheld as well as on tripod.
Others report excellent results on tripod but frustration and failure when handheld.
I believe both reports. I believe the difference is in the camera, more specifically in the mirror image PD autofocus.
Everybody seems to agree that they get excellent results from the Nikon D800(E) when shot like a mirrorless camera, under mirror lockup, using image sensor CDAF. Making image quality comparisons between different sensor resolutions must be done under similar circumstances, and using the same type of focusing. If and when you do this I believe you will find the following:
1. The main issue is not the difference in resolution.
2. The main issue is the autofocus.
A higher resolution sensor will of course resolve a higher count of line pairs including detecting a smaller degree of camera shake.
However, at the resolving powers and whole image viewing magnifications that we are talking about you will not be able to tell the difference. You will have to do some pixel peaking to see the difference.
Will a hand held 16 MP camera produce a perfectly crisp image while a hand held 36 MP camera under the same circumstances produces an out of focus looking image ? NO !
If the 36 MP Nikon D800(E) had an EVF and an image sensor CDAF comparable to the Olympus OM-D, you would get outstanding results from the D800 and leave everything else at home.
Unless you don’t like the larger and more detailed files of course.
I disagree. Looking at 100% files, there are plenty of situations in which 36MP FX will give you blurred results but the 16MP M43 won’t, assuming it’s the same photographer and he’s always consistently stable (or not):
1. The M4/3 camera has a very effective stabilizer.
2. For a given angle of view, the 36MP camera puts more pixels per degree of FOV than the 16MP mirrorless one will. That means higher resolving power, both for detail, and a higher sensitivity to camera shake at the pixel level.
3. No medium format shooter – with an even higher number of pixels per degree of FOV – would dream of shooting handheld and available light. This is a known issue with every MF shooter I’ve spoken to.
Yes, AF plays a part, but it’s not the cause resolution loss due to camera shake.
I’ve shot tens of thousands of images with both, and I’m reporting what I see. I’ve used the same technique without problems on all cameras up to 24MP – the D800E is the only one where I’m experiencing these issues.
“I’ve shot tens of thousands of images with both, and I’m reporting what I see. I’ve used the same technique without problems on all cameras up to 24MP – the D800E is the only one where I’m experiencing these issues.”
YES !! And It’s because the Nikon D800E PDAF is not accurate. It’s not because we have past a limit for practical resolution.
Wait until we finally get a 36MP mirrorless camera with CDAF on the market. It will change your mind about high resolution.
Meanwhile, let’s not badmouth high resolution based on one bad apple.
Same problem whether I’m using PDAF or using CDAF first via LV then holding it to my face for stability. It’s NOT the AF system, though that could use some serious improvements, too.
I’m getting a H4D and CFV to test today. We’ll see about high resolution…
Hi Ming,
As a D800E owner, I thank you for this helpful article.
You state that: “I define a good result as one which the image is critically sharp at 100% actual-pixels view…” Do you do your “pixel forensics” before or after sharpening the image? My in-camera setting for sharpness is set to zero so my RAW file will have some softness to it. I sharpen as a last step in my work flow. At what stage should I be doing critical sharpness evaluations?
Also, it would be helpful if you could post two side-by-side images…one in which the photo is critically sharp at 100% actual pixels and another that is close but not critically sharp in your opinion.
Thank you for your good work.
No problem. Pixel forensics before sharpening: some images are so sharp already they don’t need any unsharp masking. I do sharpening as the last step in my workflow, too, but I leave the ACR default sharpening in as you need some to counteract the bayer interpolation.
Sharpness…will be the subject of a future article. 🙂
Thank you, Ming…what is your D800E in-camera sharpening setting? Since you are doing pixel forensics on a RAW image, that is an important setting. And I really look forward to your future article on sharpness.
Actually, no, because the in-camera sharpening setting only affects the JPEG preview, NOT the raw file via ACR. It tags the setting in NX2. I turn it up to maximum (9) to make it easier to determine if I’ve nailed critical focus or not.
How does this apply to ultra high density sensors like the 41MP in the Nokia cell phone? Is that why they downsample so heavily?
I haven’t seen full size files from this thing, but I suspect that the output will look pretty horrible at 41MP most of the time. The downsampling helps with detail and dynamic range if it’s done properly. That, and I can’t see anybody sharing 41MP files on a regular basis…which is the primary aim of a cellphone camera.
So that’s why my D5000 gives me better sharpness that my D5100. I thought I was going blind.
Could be AF precision, lens quality, but yes, more pixels are definitely more demanding.
Dear Ming.
Is it therefore correct to say that the best (ie highest quality on a pixel level) downsampling size for a 36MP D800 file is 8MP, then 2MP etc?
Thanks as always!
Not necessarily, it depends very much on your downsizing algorithm. But personally with the algorithms I’ve got so far, I think those are good sizes.
This doesn’t make sense to me. Digital cameras are sampling devices of an analogue world. If you shake a D700 or D800 an equal amount, you’ll still always have more recorded detail with the D800 even if it’s blurred.
Let’s think about a scene with a really small window that has a painting you can see through it.
Even with a small but equal amount of motion blur, you’ll see more details of what that painting might look like on a 1:1 pixel level from the D800 shot than the D700 shot.
.
Jae Yoon, I agree.
Ming, thanks for the post. I am still unconvinced how detail at 100% magnification translates to an IQ disparity at typical (and equal) display sizes that *ever* favors the lower-resolution sensor.
As I’ve stated before, I cannot create a situation where I prefer the output of my D700 to my D800. Even when I’m viewing at 100%, where I may see some pixel-level blur on the D800 image (when shutter-speed constrained,) I only see lack of that same detail in the D700 shot. At equal display sizes, I don’t see greater acuity or edge sharpness with the D700 shot.
Regarding high-ISO: Even with Lightroom’s standard downsampling algorithms, my ISO 6400 D800 shots look crisper, more detailed when downsampled to 12MP than my native 12MP D700 ISO 6400 shots. Even though downsampling is not really ‘noise reduction’, it does have this perceived benefit in the final output – the finer noise grain in the D800 image translates to more pleasing prints, more ‘real’ looking output, and frankly I think it holds up to NR better when NR is applied.
I still would *love* to see a D700 vs. D800 image, same subject, same settings, in which the D700 image looks better to you. You speak with conviction on this subject, I have to believe you could save a 1000 words with just one image comparison! 🙂
.
There are two points missing here.
Firstly, I prefer the D700 where I have a lot of files to process.
Secondly, I’m finding to get an improvement at the final output size with the D800E, I have to increase the shutter speed to counteract camera shake being more obvious, which means jacking the ISO higher – so it’s never going to be a like to like comparison; it’s more like ISO 3200 on the D700 vs 6400 on the D800E. And that equalizes things.
An image comparison would be both pointless and impractical: you need to see a print, not a web size JPEG. Otherwise I’d have done it already 🙂
I also prefer my D700 when processing lots of files 🙂 But I think that’s about the extent of that preference.
Regarding your statement: “I’m finding to get an improvement at the final output size with the D800E, I have to increase the shutter speed to counteract camera shake being more obvious, which means jacking the ISO higher…” Why? How is camera shake any more obvious on the D800 when viewing or printing at the same display sizes? In the very worst scenario for the D800, those same-display-size images should look the *same*, with the D800 having an advantage in situations where you are not in any way shutter-speed constrained and there’s no camera shake introducing blur. I don’t see why you would need to increase ISO on the D800 shot compared to the D700 shot when you’re displaying or printing at the same sizes – something is not adding up… Unless this output is ultimately being judged at 100%… in which case I’d have to ask, “why”, again?
With regard to image comparisons; what I’ve done to display this concept in the fora is to take a 12MP image with the D700, a 36MP image with the D800, downsample the D800 image to 12MP and then take an equal area deep crop of each, and compare side-side. It’s in effect ‘snipping out’ an area of two images, per se, and showing the relative quality of each. It really does approximate the relative quality of the whole-image prints.
I think we’re just going to have to agree to disagree on this one. I think my position on image quality is pretty clear; I want as much as I can get, all the time. So there’s a good reason why I’m not using the D800 for everything: there is a better tool for certain situations. And really, it’s counterintuitive but you do need to see prints for this to make sense.
I understand that judging prints is the only way to properly analyze all aspect of file quality. This means that judging this on computer screens will not reveal all the differences between, say a 12, 24 and 36 mp camera. I am a bit afraid to ask a professional this question, but hobbyists like me, sometimes sadly have the money to buy these high resolution cameras. We do it for the pleasure of handling a camera high on the evolution line, and for the flexibility it offers for our not-so-important picture-taking tasks. With this perspective, would you say that there is any disadvantage using a high-resolution camera, primarly viewing the results on high-quality high-resolutions screens? I know it will be over-kill, and not reveal all quality aspect of the files, but on the other hand, given the restrictions a screen has – it could have some advantages – but no disadvantages?
Not quite – you can get screens that show up to about 4MP of resolution – the Apple Retina MBPs and 27″ Cinema displays – and you will see the difference between 12, 24 and 36, though it won’t be as obvious as with a print. This of course assumes that pixel level quality is the same for all three cameras – and you can achieve consistent results here. If you don’t print big – save yourself the computation and storage hassle, and go with something like a D4. You get the feel and responsiveness of the high end, but not the fiddliness. I don’t use my D800E for personal work, which should say something – either the OM-D or RX100 are my weapons of choice.
There is a difference between perceived acuity and real resolution – the former matters because of the way the human eye interprets detail. Something with harder edges is perceived to be sharper, even if there isn’t as much detail. On a technical level, I don’t disagree – but photography is all about perception. Furthermore, you’ve got the workflow issue to deal with – the D800Es files simply take 3x as long for me to process because of the size. I don’t have the time for that.
When I saw this article, the first thing that popped into my mind was that the 20MP on the RX100 were proving less useful than expected, for a handheld camera without a grip.
Then I read the comments and was reminded that you also have a 36MP Nikon, which does have a grip and a VF to hold against your head for stability.
So now I’m not sure which camera this article was written about. 🙂
I also recall reading that the Fuji X10 takes sharper pictures in 6MP mode than in 12MP mode. I hadn’t considered the lower number of pixels per degree of field of view as a contributing cause, but it would be very interesting (and controversial) if so.
Love your blog.
Actually, getting pixel-level sharpness on the RX100 isn’t that difficult – the stabilizer is not bad, and the number of pixels per degree of FOV isn’t as high as the D800E. I actually have a higher hit rate with the RX100, surprisingly.
The X10 looks sharper in 6MP simply because it’s downsizing. There isn’t any more resolution there.
Thanks for your reply…
So what would you say is the practical upper limit for the number of pixels per degree of FOV? Is there much room above 20MP (unless you’re using a tripod)?
I have no idea from a scientific point of view, but I think 24MP seems to be at the threshold for me at least. Anything above that requires very high shutter speeds or a tripod.
Thank you, Ming. Will keep that in mind for my next camera. Best regards.
The development of high resolution image sensors is presently ahead of the development of pixel to image processing algorithms. A lower resolution sensor will for ever leave out detail making you think that you are looking at sharpness when in reality you are looking at a straight line simplification of the subject. A higher resolution sensor will capture more detail, including accidental movement of subject and camera. However, the difference in the final image has more to do with the image processing algorithm than with the pixel count of the sensor.
Today we do not yet have pixel to image processing algorithms that give full flexibility to high resolution sensors. There is nothing that prevents the data from a 48 MP sensor to be grouped down into two by two pixels of four to produce an image essentially identical to that out of a 12 MP sensor, but with substantially less noise.
I predict that it won’t be long before you can buy a mega pixel camera representing the best of both worlds with a menu choice between; 48 MP Studio, 36 MP Landscape, 24 MP Travel or 12 MP Night.
And before that somebody might respond to the challenge of making a raw file from a Nikon D800(E) look just like one from an Olympus OM-D, but hardly the other way around.
I was hoping that the D800 would at least have a pixel-binned S-RAW option, but it turned out to be not the case. And as you point out, shooting the full size raw file and then downsizing it later doesn’t really have the same benefits to SNR or dynamic range – only pixel detail.
Yes, downsizing is a very crude routine. I believe that intelligent pixel integration with choice of emphasis on detail, sharpness, contrast, dynamic range or low noise, etc, will soon be available for post processing from people like Topaz and others.
Hello Ming,
I appreciated your article on the sharpness; however, you did not address the important issue of depth of field.
it becomes much finer with a D800 (E): based on my experience, almost 2 times less than a D200.
A focus at infinity with 35mm f11 ensures sharpness to 25 or 30 meters and the rest is not fuzzy but soft and not sharp.
With my D200, hyperfocal ensured sharpness up to 10 meters or less.
(Yes, the hyperfocal distance is not really razor sharp … but still usable, whereas in this case, whith a D800E the weakness is quickly visible)
it is comparable to a large format results but no f64 here !
Best regards
Good point – that was what I mean when saying ‘follow the DOF scale for one stop larger than you’re shooting at’ – i.e. for f8, treat as f5.6 DOF. Not perfect, but a good rule of thumb. That said, at a given print size, the perceived DOF for a 12MP and 36MP camera will look about the same (providing you’re not hitting resolution limits on the print).
Think i’ll stick with my 12mp DSLR for now 🙂
Shot well, it’ll print up to 40×60″ just fine. I’ve done many of those myself. 🙂
Thank you!. Very good report, unfortunately, they do not reach the general public.
Most buyers reduce the quality of a camera exclusively on the number of its pixels.
Thanks Winfried. Doens’t bother me, I’m writing for the thinking photographer. 🙂
Very well explained, Ming. I’ve recently bought an OM-D + 20mm (referral link used), and the only shots that were pixel-sharp from the first few batches had shutter speeds in the thousandths of a second. (and I was priding myself on my ability to handhold the G12 for up to two seconds for pixel-sharp night shots…). What would you suggest, technique-wise, to improve?
Thanks for the support! The OM-D is actually pretty forgiving because the # of pixels per degree of field of view isn’t that high compared to some other cameras, and you’ve got extended DOF compensating for any focusing errors – which in themselves will be few because it uses the imaging sensor to focus. I’ve been able to shoot the 12/2 down to 1/2 second handheld consistently; 1s sometimes. Let the IS system settle for a second or so before you shoot, and gently release the shutter. Sometimes bursts can help, too.
Psssssst. “You hear that? That was Ming cracking open a can of scientific Whoopass!!!” In all seriousness, always clear and concise, no one could have explained it better. Sounds like all that iso advantage of the D800 is a wash due to the need for increased shutter speed?
Thanks Jeff. Yes, the ISO advantage of the D800 is a wash in practical terms for a given print size, and it’s at a disadvantage if you want to take advantage of all of the increased resolution – i.e. pixel-level sharpness.
This is a great post – and something that I’ve posted about on various forums before.
My 12MP XSi was “sharp” (with good glass like the 17-55 f/2.8)… when I moved the same glass to an 18MP 7D… I was extremely underwhelmed. The pixel pitch is insane on the 7D at 4.16 micron… but it also has other issues like a strong low pass filter. I immediately noticed that I had to have a more careful technique for my landscape photography:
1. Mirror lockup became _necessary_ not just a good idea. The mirror movement would definitely cause a drop in sharpness.
2. Turning off Image Stabilization when on a tripod became _necessary_.
3. The Circle of Confusion for a 4.16 micron pixel pitch sensor is incredibly small… causing diffraction effects to show up much sooner. Shooting beyond ~f/7 would start to show diffraction effects when f/11 was fine on my XSi
4. Due to #3, proper hyperfocal focusing technique became much more necessary. Instead of just choosing f/11 and knowing that everything would be “sharp” I had to use f/6 to f/8 and use hyperfocal techniques to try to get everything in focus.
5. Better glass became more important. I was lucky that my 17-55 f/2.8 was damn sharp. But my longer lenses (and wider) weren’t up to the task… causing the need to invest…
I just recently sold my 7D (and indeed, all of my Canon gear) to switch sides and pick up a D600. I’ve really been enjoying the larger pixels in this thing! At 4.8 microns it is quite a bit more forgiving. I find myself already gravitating back toward shooting many landscapes handheld (which I never did with my 7D). Not too mention the total lack of noise (pixel size also impacts noise quite heavily… less area for averaging photon hits for each pixel), especially at low ISO.
Looks like you’ve seen this at work firsthand – the D800E is even worse because it has the highest pixel density per degree field of view of any non-medium format camera at the moment. And the pixels themselves are small, too.
Very interesting observation. In the above scenario, what is your opinion on the extent of fixing the problem with Smart Sharpen in Photoshop for motion blur in EO#1 and lens blur in EO#2?
You’re not getting any resolution back, you’re just attempting to create the illusion of acuity. Smart sharpen is the same as normal unsharp mask except with dynamic control over threshold and whether you apply it to shadows and highlights or just one of the two. Sharpening is not a solution for poor technique.
I totally agree! Thank you for your feedback.
Hi Ming,
Interesting article thank you. But I don’t think smart sharpen is the same as unsharp mask (although it may be at some settings) – it uses some form of (simplified) deconvolution when more accurate is selected – see http://www.luminous-landscape.com/forum/index.php?topic=45038.0 MadmanChan’s reply “Yes Photoshop’s Smart Sharpen is based on deconvolution (but you will need to choose the “More Accurate” option and the Lens Blur kernel for best results). Same with Camera Raw 6 and Lightroom 3 if you ramp up the Detail slider.”
I’ve oversimplified, but fundamentally all sharpening algorithms do the same thing: increase edge contrast by a user-adjsutable amount, and out to a (mostly) user-adjustable distance. They increase the perceived acuity of an edge but not generate any actual detail. The main difference I see with smart sharpen is that you can choose not to sharpen the shadows and avoid enhancing noise.
That makes perfect sense.
I’m assuming this has identical implications for smaller sensors, too, such as the 16MP OM-D vs. 12MP PEN. And also the OM-D vs. larger sensor cameras, particularly with the OM-D’s 3.7µm pixel pitch? I hope you might provide some suggestions for how to deal with this issue. Take multiple shots of a subject and always double check focus, I guess. Stick to a tripod as much as possible?
Yes and no – for starters, we’re not comparing like to like with sensors; the photosite architecture is going to be different for every sensor, and technology advancements are going to mean that smaller pixels don’t necessarily mean worse image quality. The mirrorless cameras have a huge advantage in focusing precision: you’re always focusing on the sensor plane, so resolution is always maximized. You don’t have to use a tripod all the time, but at marginal shutter speeds, the difference will definitely be felt. I find that the OM-D’s stabilizer is extremely effective, however; and this makes it easier to both frame precisely and avoid camera shake. A bit of forensics at the pixel level for an individual photographer’s images is required to determine what’s missing, though.
Thanks for the explanation. That’s really helpful.
Thank You Ming! Very well put!
Thanks Eric!