CMOS sensors make use of microlenses to direct the light into each pixel well. As seen on other Canon CMOS sensors, the sensor in the EOS-1D X uses gapless microlenses located above each photodiode to maximise the light gathering capability, no matter what angle the light arrives at the sensor from. It is the first time that gapless microlenses have been employed on a Canon full-frame sensor and they are one key piece of the puzzle that opens up the high ISO capabilities of the camera. The newly designed CMOS circuits have also been designed to capture a wider dynamic range, meaning more detail is retained in both shadows and highlights in comparison to previous cameras.

Also from your source:

if anything, smaller pixels perform somewhat better in terms of S/N ratio (while offering more resolution).

I think he’s smoking some seriously strong stuff. Please let him take a look at the D800.

Like I already said, the dynamic range not only depends on larger photosites, but also on the underlying electronics.

Okay, then explain this: the Nikon D300s has a dynamic range of 12.2 EV and a pixel pitch of 5.42µm, while the Nikon D3s has a dynamic range of 12 EV and a pixel pitch of 8.4µm. And the D300s released before the D3s, by 3 months.

So are you now going to claim that Nikon put far better electronics in a “prosumer” crop camera than they did in their top of the line pro sports model?

No, sorry, the evidence is solidly against you, and so is the underlying theory. You’ve got nothing to stand on here. If you cling to the notion that smaller pixels get you less dynamic range, you will be doing so on “religious” grounds alone.

It is easy to understand that one of the reasons digital SLRs have a larger dynamic range is that their pixels are larger. Larger pixels can collect more photons in the shadow areas before the bright ones start to overflow.

This is incorrect. From http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p2.html:

The S/N ratio is closely tied to the dynamic range (DR) of the camera. The engineering definition of dynamic range is the maximum recordable signal (the raw value at which the sensor saturates) divided by the read noise, which is the lowest recordable signal. The raw value for sensor saturation cannot be assumed to be the maximum possible raw level, and must be measured (for instance by overexposing a scene by, say, five or six stops). For instance, on a Canon 40D tested at ISO 100, the sensor saturates at raw level 13825 rather than the maximum 14-bit value of 16383; the max signal is thus 13825-1024=12801 after subtracting the bias offset written into the raw data. The read noise at this ISO is 5.5 ADU, and so the dynamic range is 12801/5.5=2327, or 11.2 stops.

and from http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html:

Bottom line: Among the important measures of image quality are signal-to-noise ratio of the capture process, and resolution. It was shown that for fixed sensor format, the light collection efficiency per unit area is essentially independent of pixel size, over a huge range of pixel sizes from 2 microns to over 8 microns, and is therefore independent of the number of megapixels. Noise performance per unit area was seen to be only weakly dependent on pixel size. The S/N ratio per unit area is much the same over a wide range of pixel sizes. There is an advantage to big pixels in low light (high ISO) applications, where read noise is an important detractor from image quality, and big pixels currently have lower read noise than aggregations of small pixels of equal area. For low ISO applications, the situation is reversed in current implementations — if anything, smaller pixels perform somewhat better in terms of S/N ratio (while offering more resolution). A further exploration of these issues can be found on the supplemental page. Rather than having strong dependence on the pixel size, the noise performance instead depends quite strongly on sensor size — bigger sensors yield higher quality images, by capturing more signal (photons).

(emphasis mine)

Taken together, the dynamic range is, basically, not a function of pixel pitch, unless you insist on measuring the S/N of only a single pixel. This is because the S/N of an area of small pixels is the same as the S/N of a single pixel occupying the same area and the smaller pixels will reach saturation at the same time that the larger pixel will, because the light is divided evenly across the smaller pixels.

The aforementioned source is my primary source for this stuff. Dispute it if you want, but you’ll be arguing against someone who studies this kind of thing for a living.

5DmkII? I like crop for safari I’m going on and wildlife etc. FF still a good choice?

Really, really don’t want to go Nikon..

You’re wrong, dynamic range DOES depend a lot on the size of the photosites. From here:

Then explain how the Canon Powshot S95, with a pixel pitch of 2.03 µm, gets a dynamic range of 11.3 EV, just 0.2 EV behind the 60D which was released at the same time and has a pixel pitch of 4.29 µm, more than twice the diameter (which means more than 4 times the area). The more recent Powershot S100, with an even smaller pixel pitch of 1.8 µm, matches the 60D’s dynamic range.

No, sorry, the evidence is solidly against you on this.

If there is any relationship at all, it is so small as to be of no real consequence.

It is easy to understand that one of the reasons digital SLRs have a larger dynamic range is that their pixels are larger. Larger pixels can collect more photons in the shadow areas before the bright ones start to overflow.

Notice how the problems on the D800 are especially in the darker parts of the images. This was also a big issue on the 5D Mark II which I also blogged about. There is no hope for the sensor in the D800. Software can’t fix these issues. In the comparison of the 1D3 and 1Ds3, there may also be other factors such as the design of the sensor that influence dynamic range. For example, if you read about the sensor in the 1DX, Canon not only made the photosites much bigger and efficient for light gathering but they also improved the electronics resulting in better dynamic range.

Sensors are analog devices, not digital. They build an electrical charge in the pixels as the light hits them. Smaller pixels merely collect fewer photons each during the exposure, and thus build a smaller charge during the exposure. The random variation in charge between pixels results in noise. If the size of the charge is greater, the random variation becomes less as a percentage of the charge, and the end result is less perceived noise.

The banding in that D800 shot is not random noise! it is a consistent error in the color representation (which is why it’s banding, and not color blotching or something that would result from color noise reduction), and therefore it cannot be from the sensor directly. If it’s in hardware at all, it will be in the analog to digital converter, or maybe the hardware noise reduction unit, if the camera even has such a thing. Much more likely, though, the problem is somewhere in the image processing firmware in the camera, or in the image processing software that was used to convert the raw image files to JPEG.

Dynamic range is the ratio of the highest light intensity that can be recorded to the lowest light intensity that can be recorded. Dynamic range is not a function of pixel size. If it were, then the 1D mark 3 would lead the 1Ds mark 3 in dynamic range. Instead, it lags it by 0.3 EV in dynamic range (the 1D3 has 11.7 EV of dynamic range, while the 1Ds3 has 12. Source: http://www.dxomark.com). The 1D3 has larger pixels than the 1Ds3: 7.2µm versus 6.3µm. This is an existence proof that dynamic range is unrelated to pixel pitch.

You should try out mRAW just to see what it does to noise. You may be pleasantly surprised.

That said, it does not come from the sensor, but from the way the image is processed. Either Nikon has a problem with their A/D converters (unlikely, but possible), or they have a problem in their image processing pipeline after the fact (much more likely). Looks like Nikon has some work to do on that front.

We’ll see how things really are once both cameras have been released. JPEGs from the camera will show you the minimum of what the camera is capable of, because they involve the image processing pipeline of the camera. Few would bother with processing raw images themselves if the in-camera JPEG converter were better than anything that can be done on the computer. Processing raws on the computer has much greater potential, in large part because the time constraints that cameras are placed under do not exist there, but also because you have a much wider range of postprocessing options at your disposal.

mRAW won’t take care of anything other than noise, because it’s just like properly downsizing an image, with all the attendant reduction in noise that comes with it. Have you shot in mRAW to see for yourself what the results are?

My point is only that at least one manufacturer is going in a direction opposite of what you apparently expected. What kind of image quality that results in will depend on how the file is postprocessed and what the target size is.

Some people (landscapers, especially) are going to love the D800. Others, not so much. Seems to me the biggest problem is that Nikon doesn’t have the equivalent of Canon’s mRAW, so the camera can’t shoot lower resolution raw files.

By the way, I don’t think you ever responded when I mentioned shooting the 7D in mRAW. Did you try it? If so, what did you think?

So … what were you saying, again, about manufacturers scaling back the resolutions of their sensors because they recognize the “truth” of higher resolutions yielding lower image quality?

I rest my case.

That said, I should note that downsizing has to be done properly when the original resolution and target resolution are relatively close, as they are in the Nex-7 versus Nex-5n comparison. That comparison would, if the per-pixel noise were the same between both prior to downsizing, yield a little more than half a stop of improvement in noise as a result of downsizing from 24mp to 16mp.

The reason it’s important to be careful of how one goes about downsizing is that most downsizing algorithms, such as bicubic, tend to attempt to preserve detail. In doing so, they will also tend to preserve noise. In my experience, when downsizing like that, one has to first perform a gaussian blurring pass over the original prior to downsizing. Selection of the radius is, of course, critical in that case.

I don’t know what kind of downsizing algorithm was used for that comparison. I expect it was likely to be bicubic, which is the default resizing method used by Photoshop. It would be very interesting to experiment with both of the original images.

Regardless, assuming that the noise difference isn’t entirely accounted for by the downsizing methodology, the comparison illustrates that, indeed, sometimes more megapixels does yield more noise at a given resolution. Everything depends on the implementation.

The 7D’s sensor does not appear to be noisier on a per-physical-area basis than is the 5D2′s sensor. The reason I say that is that the 7D is behind the 5D2 by about a stop in terms of high ISO noise. Which is to say, the noise signature of the 7D at, say, ISO 3200, is about the same as the 5D2′s at ISO 6400. This is entirely accounted for by the sensor size difference alone, which yields a little over 1 and 1/3 stops advantage in favor of full frame. This means that if you were to take the 7D’s sensor and scale it up to full frame size (which would get you a 46 megapixel sensor), the noise signature of images from it which are properly downsized to match the 5D2′s resolution would be the same as the 5D2′s noise signature.

It’ll be interesting to see if the reversal of direction Canon is taking with the 1DX is universal or not. My prediction is that the next generation of other camera bodies from Canon will at the very least retain the current resolutions, if not increase them. I suspect the decrease we see in the 1DX is an anomaly and not part of a larger trend.

I think this is a particularly useful comparison because of the proximity these cameras have to each other in terms of release date, company and technology.

http://www.stevehuffphoto.com/2011/10/23/the-sony-nex-7-digital-camera-review-by-steve-huff/

Everything there is to say has already been said. If you can’t understand based on the info I’ve given so far, then you won’t and I can’t convince you. Apparently not even Nikon and Canon can. Just stick to what you think if you want to. I’m done with this discussion.

The problem is that you haven’t provided any information, only assertions followed by argument by appeal to authority (which is a logical fallacy). Those are not the same thing.

But I agree, it appears we cannot convince each other of our respective positions. That’s fine. That does happen sometimes. I’ve given it my best, and you’ve hopefully given it your best.

So we will agree to disagree! Interested readers can take away the arguments presented and form their own conclusions.

Here’s hoping Canon produces a camera that meets your image quality needs. The 1DX certainly looks like it’ll be an excellent all-rounder.

If you haven’t done so yet, try shooting your 7D in mRAW or sRAW. You might like the results better than the results you get at full resolution.

LOL dude, tell dpreview, and especially DXOMark not to compare noise at 100%.

Has it occurred to you to ask why they compare noise at 100%? It’s simple: because they’re giving you the raw information that you can derive your own conclusions from. It’s not their fault if you choose to derive your conclusions using faulty reasoning, or reasoning which fails to account for your own use case.

That’s how noise should be compared.

So why aren’t you shooting your 7D in sRAW (4.5 megapixels), then?

Let’s get down to brass tacks.

Why is that how noise should be compared? Don’t just make the assertion, back it up with reasoning to justify it.

All you’ve done thus far is to assert that 100% is how it should be compared. I have yet to see you say why it should be compared that way. Conversely, I have given reasons for comparing it in other ways.

Asserting something without giving the reasoning behind it means that you either don’t truly understand what you’re asserting, or that it’s a belief akin to a religious belief, i.e. something that you simply take on fatih.

And the 7D records marginally more detail compared to the 40D, which gets lost once you do some luminance noise reduction which is badly needed on the 7D even at ISO100.

“Badly needed?” Under what use cases? What kind of noise reduction algorithms are you using? What RAW postprocessor are you using? Are you using a masking layer to do sharpening, or are you merely doing a sharpening pass throughout the frame?

How much actual detail the 7D records depends greatly on the lens you’re using. It’s very demanding of the lens. Put a really good lens on it and the amount of detail recorded is quite a lot more than what the 40D can record. I know this through experience. I’ve taken landscape shots that show detail that the 40D simply would never be able to capture, because it just doesn’t have enough resolution. How do I know that? Because the detail is at the pixel level at 100%, and it is lost when downsizing to 10 megapixels.

So what lenses have you been using on your 7D to do this comparison against the 40D?

Then an upscaled 40D image is practically identical with regards to resolution. I take pics daily with the 7D now and honestly, you just get more blur in fine detail such as trees in the distance etc. with lots of noise.

What RAW converter are you using? Like I said, there’s a green channel imbalance in the 7D, and because of that, most RAW converters will average the green channels together. That results in loss of detail, most especially in trees and such. If you use Lightroom or ACR 6.x, the RAW converter will average the green channels together only in the areas that don’t have detail in them.

Once you do noise reduction all that is left are blobs of color that should resemble trees because along with the noise, it also removes detail.

You must be using really lousy noise reduction. The noise reduction in Lightroom allows you to specify the amount of detail to retain during noise reduction, so that the areas that are reduced in noise are the ones that don’t carry detail. Between that and the masking feature in sharpening, you can retain the detail while reducing the noise.

Both Canon and especially Nikon now admit that there’s more to image quality than resolution and that low noise is an important feature to have.

Note that I have not been arguing that resolution is the only metric behind image quality, only that it is something that can be traded for image quality. It allows you to decide where on the noise versus detail curve you want to be. Different applications can tolerate different amounts of noise.

Again, why would you want less flexibility instead of more? Less flexibility is what you’re arguing for here.

And that includes scaling back their sensor resolution in newer camera models. You can choose to ignore it if you want to.

I’ve already given the reasons they’re scaling back. Image quality is something the marketing guys are pushing as an additional benefit in order to justify the resolution reduction. Watch that argument suddenly disappear if/when they release their next batch of high resolution cameras targeted at landscape and portrait shooters, who place great value on resolution.

The cameras in question are targeted at sports shooters. Maximum resolution is not what such photographers are interested in.

So why is Canon ditching the 1Ds series, and merging it and the 1D series into the 1DX? Simple: because the price they’d charge for the next 1Ds series camera would put it squarely in medium format territory, where it would be competing with cameras with larger sensors, and that would put it at a disadvantage in terms of actual recorded detail.

it is nearly certain that the next iteration of the 5D will have higher resolution, given the 5D series’ target market.

Dude, you keep going on and on repeating faulty reasoning that even Canon used before and are now backpedaling on. If resolution is so important, why did they go back in megapixels in the G11 and now the 1DX??

If my reasoning is faulty, then you should be able to point to the specific piece of logic that is faulty and refute it with something other than mere assertion. You should be able to refute it with objective evidence, and not merely the marketing decisions of a couple of camera manufacturers regarding their top of the line sports models. You haven’t done so. Instead, you have simply repeated assertions without substantiating them. I have provided evidence backing my assertions.

You’re using marketing material to back your position. That’s not a terribly strong way to back your position.

See that? Why is Nikon sticking to 10MP if resolution is so important? The fact of the matter is that resolution increase is useless with all the noise being introduced. Canon finally saw this themselves after years of me and many others complaining.

You forgot to highlight the real reason Nikon is dropping back to 10 MP. Allow me:

Its not a megapixel race anymore but instead a development of quality and the 10mp sensor in the Nikon 1 series has been specially selected to enable high speed.

(emphasis mine)

Nikon is going to attempt to compete against Canon in terms of speed (frames per second), so they selected a target resolution that allows them to do so. Like I said previously, speed is one of the significant reasons Canon dropped the resolution of the 1DX. But tell me: if they really thought megapixels didn’t matter, then why didn’t Canon drop back to 10 megapixels as well? Why not 6 megapixels? Why not 4? If resolution doesn’t matter then why aren’t the camera manufacturers using use the biggest photosites they can build, and thus the lowest resolutions they can produce?

No. Resolution does matter. If it didn’t, we’d all be shooting with 1 megapixel cameras.

And if Nikon’s assertion that the megapixel race really ended years ago, then explain the existence of the D3x, and why Nikon continues to sell it.

If resolution doesn’t matter, then explain the existence of ultra high resolution medium format cameras.

Mentioning print sizes of 20×30 also doesn’t make sense. I can print a 40D image at 20×30 and a 7D image and you’ll never be able to tell which one is which. It’s all about how you upsize a good quality image.

That’s true as far as it goes, but upsizing cannot recreate detail that has not been captured. It can introduce false detail, but that’s not the same thing.

To whatever degree you can upsize the 40D image, you can also upsize the 7D image and wind up with an even larger print. And to refute your assertion that the quality is the same, explain this bit in Imaging Resource’s evaluation of 40D prints:

Excellent print quality at 13×19 inches, a little softer at 16×20.

If the 40D prints weren’t becoming resolution limited at 13×19 inches, they wouldn’t be getting soft at 16×20, which is already below the print size at which the 7D remains crisp.

The choice of upscaling method makes a difference, of course, but as I said, it cannot recreate true detail that isn’t there, and anything that it can do to a lower resolution image, it can also do to a higher resolution image. If it can extend the size one can print the lower resolution image, it can also extend the size one can print the higher resolution image.

The fact of the matter is that at ISO 100, the 7D records more detail, at a noise level that is below what shows up in print. The 5D2 most certainly does.

Now, maybe you have no need to print large enough to see the difference, but you are not the entire world.

I’ll take a clean 18MP image any day over a noisy 32MP image.

Then downsize the 32MP image to 18MP and get a clean 18MP image. The 32MP image gives you that option. You can choose the level of cleanliness you want if you start with a higher resolution image. With a lower resolution image, you’re stuck with what you have.

It’s not like downsizing an image is a hard thing to do. If you’re using lightroom, it will automatically downsize to whatever resolution you choose upon export.

Honestly, I just don’t get why you’d choose less flexibility over more flexibility when the image quality is going to be identical in terms of noise at the same (lower) resolution.

You can go on if you want to with this, but both Nikon and Canon are now showing you what is important.

No, they’re showing you what they believe the market thinks is important. They’re responding to the mistaken belief that many in the market share with you, that lower resolution is somehow “better”, based on the mistaken notion that noise should always be compared using 100% shots on the screen regardless of any disparity in resolution.

Well, that’s fine. Canon and Nikon are giving you what you want. I would suggest, then, that you should take whatever image you want to print, downsize it to 1 megapixel to get rid of all noise (because the 40D does have noise at ISO 100), and then print that to whatever large target size you want. If you can print a “better” quality lower resolution shot to the same size that you can print a “worse” quality higher resolution shot and get the same results, then downsizing to get rid of all the noise prior to printing will get you optimal results, right?

Yeah, right. Good luck with that.

You’re free, of course, to retain your religious belief that the only way to compare noise is with 100% shots. I hope Canon offers a 900×600 pixel full frame camera for sale just for you. Its image quality will be out of this world, with noise free images all the way up to ISO 51200, and acceptable noise all the way up to ISO 819200. Ask yourself why you wouldn’t want such a camera, and maybe the answer you come up with will give you insight into why the arguments you’ve been using here are fundamentally incorrect.

However, Nikon believe that the megapixel race ended years ago with the Nikon D3s. Its not a megapixel race anymore but instead a development of quality and the 10mp sensor in the Nikon 1 series has been specially selected to enable high speed. The Nikon D3s is the prime example of what you can do with a sensor and the 12mp Nikon D3s is still the market leader.

See that? Why is Nikon sticking to 10MP if resolution is so important? The fact of the matter is that resolution increase is useless with all the noise being introduced. Canon finally saw this themselves after years of me and many others complaining.

Mentioning print sizes of 20×30 also doesn’t make sense. I can print a 40D image at 20×30 and a 7D image and you’ll never be able to tell which one is which. It’s all about how you upsize a good quality image. Even Canon mentioned in their release of the 1DX that the 1DX 18MP images are so clean that you can upsize them a lot and make up for the smaller size. THAT is what image quality will do for you!! I’ll take a clean 18MP image any day over a noisy 32MP image.

You can go on if you want to with this, but both Nikon and Canon are now showing you what is important.

It would be nice if it were possible to edit these messages after the fact…

Many people do what I do to compare noise, namely blowing up the image to view at 100%, because that’s the way to compare noise effectively.

No, it’s not the way to compare noise effectively, it’s the easiest way to compare noise. The easiest way to compare noise is not necessarily the best, just as the easiest way to take the picture (“green box mode”) is not necessarily the best.

Again I ask, because you have not yet answered, this question: what use case demands that noise be compared at 100% magnification independent of target resolution?

The most effective way to compare noise will have to account for the way the images will be used.

Consider printing. Here’s what Imaging Resource says about the 7D with respect to printing (source):

Exceptional print quality, good color, very sharp 20×30 inch prints from both JPEG and RAW. Exceptional print quality at all high ISOs as well.

(emphasis mine)

And here’s what that very same source says about the 40D (source):

Excellent print quality at 13×19 inches, a little softer at 16×20. Up to ISO 800 are still great at 13×19 inches, and ISO 1,600 looks good at 8×10.

(again, emphasis mine)

Notice that the maximum print size from the 7D is significantly larger than for the 40D, while still maintaining “exceptional” print quality. That is the result of the 7D having greater resolution, even if it has higher ISO 100 noise at 100% magnification on the screen.

Not downsizing images. Otherwise the resolution increase is useless.

Useless to you is not the same as useless to everyone. I already outlined how the tradeoff can be useful, and relatively objective sources such as Imaging Resource implicitly agree. Printing large is one example. There is a noise threshold below which one cannot see the noise in print. Any noise that comes in below that threshold is going to be perfectly acceptable for that purpose. From what people who have printed ISO 100 shots from the 7D have said, they are unable to see the noise in print at print sizes where the resolution advantage actually proves useful, else the print quality would degrade at those print sizes. To them, the extra resolution is useful even though it may come at the expense of extra noise on the monitor.

Apparently you have a different opinion and that’s fine, but even Canon now has to start backpedaling because many photographers complained about the noise in images.

Yes, that’s part of the reason Canon dropped to 18 megapixels. And why did those photographers complain about the noise? Because they were viewing the shots at 100% on the monitor! Again, that’s fine as far as it goes, but it ignores the very question I asked earlier: what’s the use case? Without the answer to that, it’s impossible to know whether or not viewing the image at 100% is a reasonable way to evaluate noise levels.

This is why they went down in resolution also with the G11 which I mentioned in my post. People want better image quality not senseless megapixels.

No, the people in question want better image quality at 100% magnification, regardless of whether or not that magnification is actually useful to them. But in any case, as I said, in the case of the 7D, if you want better image quality at the pixel level than the 40D provides, you need only shoot in mRAW, which gets you better image quality at 10 megapixels than does the 40D, with no additional effort on the part of the photographer.

Extra resolution gets you more flexibility. It’s as simple as that. You can make use of that flexibility, or you can mischaracterize it as a loss of image quality. Many have chosen to mischaracterize it because they either haven’t yet learned about or refuse to understand the tradeoffs even when those tradeoffs have been explained thoroughly. That’s fine. Canon is most certainly willing to accommodate them! After all, it’s cheaper to make a lower resolution sensor than a higher resolution one. But I would suggest, then, that based on your own reasoning, the 5D classic has better image quality than the 5D2, because at 100% view at ISO 100, there’s less noise in the 5D classic shots than in the 5D2 shots independent of the deep shadow pattern noise problem. You can call that a step backwards in image quality if you want, but people who actually shoot with the 5D2 and who have shot with the 5D classic previously will vehemently disagree with you, except with respect to the problem of pattern noise in the deep shadows.

The actual use case is everything when evaluating a piece of equipment. You don’t evaluate the speed of a pickup truck in comparison with that of a sports car and declare the sports car superior for all purposes, because their typical use cases are very different. Similarly, you cannot properly evaluate the image quality of a high resolution camera in comparison with a lower resolution camera without first knowing the use to which its images will be put. And that leads right back to the question I’ve been asking all along, to which you’ve thus far been silent: what is the use case which demands that shots from different cameras be compared at 100% for the purpose of evaluating the noise?

Answer that, and we can get further with this discussion. Fail to answer it, and it will be difficult to conclude anything other than that you have a religious attachment to evaluating noise at 100%.

For the record, I once thought as you do, until I started to really think things through, and learned about how noise and resolution relate to each other.