Perhaps I'm missing something. If you can sharpen up a diffraction limited file so that it (almost) matches a non diffraction limited file, does that not mean that if you were to apply the same sharpening to the non diffraction limited file you would pull even further ahead? In other words the resolving ability of the sensor should be based on applying the same changes to each file and the percentage advantage would always go to the non diffraction limited file. If you want 21 megapixels worth of resolution from the 5D mkII then no matter how much you play with a diffraction limited file, it won't have the 21 megapixel goodness achievable by the file shot at f5.6 because whatever you can do to improve the f16 file, you can do that to the f5.6 file as well and the f5.6 file will always be better by specific amount. That also means that your sharpened f16 file is still not 21 megapixels. It will always be 'limited'.

I'm very far from being a scientific mind so please explain to me if I'm assuming wrong. I hear many things such as a D700 file can be sharpened up to match a 5D file, downsizing reduces noise, just never actually been able to make them happen in the real world. I've never been able to produce detail or sharpness from a diffraction limited file to match a non limited file also processed to its best ability. One of the reasons I've stuck with my 5D as much of my work is f16+, I stitch for the added resolution rather than use a higher res camera and lose the detail to DOF needs.

I also find it interesting that no one seems to be pointing out that the 50D is very much iso limited due to its higher level of megapixels (noise/banding over iso 400) and the D3X and A900 seem to hit a barrier at higher iso's compared to their bigger pixeled bretheren. Of course the next generation will be better as technology advances but that technology will also apply to the bigger pixels relatively. The D3's sensor is better than the D3X and the D3s will be better than the D3Xs. The technology advances make 'pushing edge' sensors with limitations but allow 'normal' sensors with very few limitations. My point is that although later technology seems to improve on what was once barely acceptable at the pushing edge limits of sensor design, soon enough it might be time to stop pushing and just start consolidating.

Lets admit it, we're more likely to get a 5D mkIII with 30 megapixels than an AF that would be worthy of pro use of the 12 megapixels of the 5D mkI or a 5D mkIII with a 21 megapixel chip that has noise properties of a D3. Methinks it's a shame.

Maybe not. Speaking as a mathematician, the point seems to be that diffraction takes an image with a given amount of detail and "convolves" the image into a less sharp image ... but that convolution is mathematically invertible (or close), so fancy algorithms like "deconvolution" can "unscramble the egg", more or less restoring the pre-diffraction image, restoring resolution to the limits set by lens aberrations, OOF effects, camera and subject motion, etc. For the mathematically inclined, the Backward Heat Equation is a similar tool. But further sharpening cannot undo those other resolution limits. For example, OOF effects are a kind of convolution that is not invertible unless the subject is all at a known distance from the camera, so OOF blurring cannot be undone by deconvolution on a single image.

I am for "comfortably enough resolution for almost everyone" plus cost/benefit trade-offs setting the ultimate pixel count limits, not the mixture of physics and high level mathematics being discussed lately. And for all I can tell, 60MP Bayer might already be there or a bit beyond.

I agree, for me there is just no benefit to a 100mp DSLR or really anything above the current crop of DSLRs. Bigger sensors mean bigger files, so you need more/bigger memory cards and more/bigger hard drives to store the data on. If you are in the field on a trip you have more GBs to manage.

We can build cars that go 200-300mph or faster, yet hardly anyone buys them because they are just not needed for most users. That same idea applies to photography.

Isn't that because they are rather expensive?

200 MPH cars ARE very expensive, but cars in the 130-150 mph range are fairly standard (any "sporty" model, plus V6 and V8 versions of many family sedans) - which is a classic wasted feature - even if every police car in the country had two flat tires (eliminating speed traps), the roads (at least in the US) won't safely support speeds much over 80-85 in very many places. The same is true of a 100 mp DSLR - if a camera in the 24-30 mp range prints 24x36 inches, how many people need to print 48x72 if quality scales with pixel count?

-Dan

Perhaps Moore's law will next manifest itself in 2" x 3" sensors?
2" x 3" sensor, 13 stops of DR, 100 mpix, 4"x 6" screen with live view, front movements.
Marc

Indeed. Pushing, to the limit, some of the arguments we've seen can lead to some fancy extremes. :-) And yes, about the 50D... the issue of ccd/cmos well capacity, quantum efficiency, thermal noise at normal operating temperatures haven't been really explored in this recent salvo of opinions. Smaller wells will definitely have a more limited electron capacity and that will have have a significant impact on DR, achievable contrast, etc...

Last but not least, there is a letter in last week's Nature about nanoscale self assembled spherical lenses (http://www.nature.com/nature/journal/v460/n7254/full/nature08173.html) that could be useful for super resolution microscopy. This letter is interesting in the sense that those academics, obviously experts in their fields, still refer to Abbe, Rayleigh and Sparrow's resolution limits. While they do manage to break the barrier, it is hard to see how it could be used in normal photography.

I doubt the lenses required to produce a correct 3" image circle could fit in a small package though... ;-)

There are some tiny lenses that have been doing it for a century. They're called large format lenses.

Yes, the lenses themselves don't necessarily need to be huge (in some cases, a pinhole will do...) but they won't fit (at least meaningfully) in a small package. CCD/CMOS are also more sensitive than film to the direction/angle of the incoming photons. I am ready to be wrong on this, but if I am, someone please tell Leica. Then, of course, small lenses are a bit on the slow side aren't they? I find large format photography to be fascinating but if there is a domain of photography that has barely been touched by the version of Moore's law we've been talking about, it has my vote. Speed hasn't increased that much, cost hasn't been going down, etc...

I quoted Mr Merklinger:
" A 40 MP DSLR would generate approximately a 60 MB raw file, might well out-resolve all existing zoom lenses in the corners, leaving full utility only with a few primes, and would almost certainly be so sensitive to camera shake that it only achieved its full resolution on a sturdy tripod (even a 24 MP DSLR is almost tripod-dedicated, the hand holding speed on the D3x seems to be around 1/250 second)."

Everyone knows the shutter speed relation between focal lenght, light condition, ISO but I have never heard that the pixels density of a sensor would influence the shutter speed. I will certainly test that assumption with my 5D MK II, in the meantime could someone give some links or explanations about that. The Fudji Velvia was one of the finest films on the market i wonder how mpx it would be associate with? Just to make a comparison.

thanks
marc-andre cote
montreal

Everyone knows the shutter speed relation between focal lenght, light condition, ISO but I have never heard that the pixels density of a sensor would influence the shutter speed. I will certainly test that assumption with my 5D MK II, in the meantime could someone give some links or explanations about that. The Fudji Velvia was one of the finest films on the market i wonder how mpx it would be associate with? Just to make a comparison.

thanks
marc-andre cote
montreal
[/quote]


If the images lateral movement at your cameras sensor is 6 microns in 1/125th of a second and your pixel pitch is 6 microns you just blurred your image 1 pixel if the shutter speed is 1/250th/sec or your pixels were 12 microns your blur would be half a pixel. So blur is worse the smaller the pixel, so are the effect of lens imperfections impurities in the glass etc. So I hope the future brings better pixels not just more pixels.
Marc

Thank you Marc for the explanation,

I'll test that in the real world with real print. A 23 x 16 inch at 240 dpi from a 21 mpx gives a 30% magnification in photoshop.
I suspect that the blur is visible at a 100% magnification which fortunatly nobody print this size.

marc-andre

Pierre and all

I think the best answer to this was given by Harold Merklinger - he mentioned superexposure and electronic filtering.
Now lets take a look what the industry already does.
Pentax has launched a DSLR which can do HDR automatically.
Sony´s Alpha has a moving chip that would perfectly fit for superexposure tasks, there could even be specific resolution progs with
diagonal offsets or symmetric doubling or whatever needed to be shot in a reasonable short timeframe.
Computing power is still getting cheaper a lot almost every year and more important- intelligent software and programming optimization
will further increase intelligence and shere numbercrunch potential in cameras.

We are just at infancy of digital image taking- expect a lot more to come. Don´t think analogue and traditional physical optics - think digital, networked and crosstechnology.

Greetings from Munich
Stefan Steib

Some little more "food" for thoughts

http://www.physics.bc.edu/Deptsite/news/padilla2.shtml
http://www.nature.com/nphoton/journal/v3/n...ton.2009.3.html
and a third link about that in German
http://pressetext.de/news/090804002/tarnka...este-simuliert/

and a similar subject
http://www.ee.princeton.edu/people/Fleischer.php
and here an according link in german
http://pressetext.de/news/090504002/neue-l...-im-weitwinkel/

Greetings from Munich
Stefan Steib

In theory yes, but whether actual products emerge will depend on commercial decisions driving where Nikon wants to focus its technological investements.

The only domain where the D3 sensor is better than the D3x is high iso noise. On all other accounts, the d3x is better - and that includes DR, which shows that technological focus is more important than basic physical laws. Nikon decided that DR was key for the D3x, and they managed to get smaller pixels but much better DR still.

It was not obvious at all that Nikon would produce the D3 with only 12 MP in the first place, it took tremendous corporate bravery to stop the megapixel race and focus on better pixels, a move that no company had done before on a large scale for a live or die kind of product.

So I am really not sure that they will dare to do it a second time... bigger pixel might be a dying breeze, or to put it otherwise, I am not sure that the major investement in terms of technology will still focus on bigger pixels. Few if any of us have enough understanding of the very complex underlying technology to figure out whether optimizing a design/process for a 30MP FX sensor will have any actual benefits for a 18MP sensor.

My bet is that a very optimized 30MP sensor where Nikon will have invested 100M$ will have better high ISO noise than a 18MP sensor on which Nikon would only have invested 10M$, even if they are released the same day, because the technology required for the 30MP might not readily apply to the 18MP sensor.

Cheers,
Bernard

Hi,

D3 and D3X sensors are so far I understand different projects. To my understanding development projects stretch over several years so there used to be several design at different stages of the development process at the same time. Regarding the D3 / D3X, I'd suggest that the D3 was a very smart move by Nikon. For one thing, the D3X sensor was probably a couple of months away. But, more importantly, the D3 was focused on those who needed to capture the picture, reporters. Nikon was extremely succesfull with replacing a lot of white lenses with black ones at the 2008 Olympics.

Canon used to go the same way, 1D for photojournalist and 1Ds for studio/landscape. 1D uses smaller sensor with relatively large pixels and has high FPS.

Regarding the D3X Nikon has found an extraordinary rabbit in their hat, I don't really understand how that very good DR seen by you, but also measured by DxO is achieved. To my understanding DR is dependent on "well size", the number of electrons a pixel can hold, and the noise in the sensor. It seems to me that the major contributor, normally, is photonic noise (random variation of the number of photons captured). To me it seems that Nikon could do very little about either.

Best regards
Erik

Perhaps that DR is not from better photo-sites but from in camera shadow noise reduction to the RAW data? If I reduce the shadow noise on a test target wouldn't that increase the tested DR?
Marc

This Nikon microscopy page is interesting

http://www.microscopyu.com/tutorials/java/.../signaltonoise/

while it doesn't apply directly to photography, it shows that the main factor influencing the SNR is Quantum Efficiency (provided the well is big enough and responds lineraly enough of course). The raw QE of documented sensors (for example the Kodak ones used in amateur or pro astronomical CCDs) ranges from from 20% to above 90%. Buil calculates a QE of 5 to 8% for the 5D. This doesn't necessarily tells much about the underlying sensor because there are so many different layers in front of it, but it certainly leaves for improvement in the whole chain.

BTW, I thought intuitively that read noise made a bigger difference, and indeed some Nikon cameras apparently have significantly lower read noise than Canon ones, but that isn't enough to explain everything...

Hi,

DxO checks for spatial filtration on RAW by using autocorrelation. This is described on their pages. DxO has also a somewhat odd definition of DR. To calculate DR you need some definition of acceptable signal. DxO uses S/N = 1 as floor. But that is equal for all. One guess on these forums was that Nikon would essentially make two readouts, with two preamp settings and combine into one raw, this is something that can be done with CMOS. It would explain a couple of things:

- Achieving 14 bits with on chip ADCs which probably are 12 bit.
- While 14 bit readout is so slow

But, according to real experts, this would leave tell tale signs in the raw-file which cannot be seem.

Best regards
Erik

In the discussions in The Luminous Landscape about this subject there are two different points of view.

The first one (1) considers the pixel as the minimum circle of confussion. Systems should be evaluated considering the pixel size (and 100% magnification on screen) as the point of reference. This line of reasoning can be seen in the discussions about diffraction limit and pixel resolution. The only problem here is how to relate pixel size and Airy disk diameter, because it depends on the nature of the real detail to be photographed. Isolate points have different requirements than line pairs, etc.

The second one (2) considers the circle of confusion (depending on human visual accuity, viewing distance, size of the print, etc.). This line of rasoning puts the limits below that of the first way of thinking. The point here is the final output (the print), and not the matrix input (the sensor, or the RAW file), is that counts.

The two arguments are valid to some extent.

Where is the limit in total number of pixels? Is it determined by diffraction limits (1) or by print size/resolution convenience (2)?

I think the first approach (1) is more rigorous for technical discussion, but the second approach (2) can be the relevant one for market discussion.

The idea is, even if you can have a 100Mp camera not limited by diffraction (thanks to a highly corrected lens), what is the practical utility of so high resolution? I am not thinking on special applications, but on the typical professional applications of photography, with a printed output (magazines, books) or published online (internet). Considering the maximum size of the publication, and the maximum resolution (pixel per milimeter), you have a maximum useful resolution (with a margin for crops, etc.). A way for improvement would be more information per pixel (full color information instead of an interpolated result, more bit depth, etc.), but no more pixels. The only practical utility of 100Mp would be the use of 4 bayer (input) pixels for getting 1 output pixel...

There are technical limits to the increase in resolution (diffraction is one of them), but I think there are also market/practical limits. Even if computers get more and more powerful and storage space gets cheaper, the practical utility of larger and larger pictures is doubtful.