The rift between the vocal "ISOers" and the "MPers"...

[torger]

What type of improvement can be expected, knowing that I am talking about a phantom body, given that an D800 equivalent of the rumored 5D3 might have 36 MP. Compared to a still to be released 1DX, what is the compromise? Will 36 MP on 5D3 yield one stop high ISO improvement at RAW over 5Dii applying the new sensor technique? Or is that unlikely combined with such an MP number?

I am no tech, therefore the question.

The big compromise will be that a 36 MP will not have decent ISO 25600 and 51200. However, the ISOs it does provide it will probably do rather well. Usable ultra high ISOs is probably also a thing that Canon would want to reserve for the flagship, so I don't think we will se more than 25600 in any case.

36 MP fullframe is exactly the same pixel density as the fairly new Nikon D7000, so that can give us a lead of what performance can be expected. Per pixel the small D7000 pixels perform about as good as the larger 5Dmk2 pixels, at low ISO the D7000 has considerably better noise performance though.

Based on this I would think that a 36 MP 5Dmk3 would yield a ~2/3 stop ISO improvement over current 5Dmk2, and substantial noise improvement at base ISO.

 

[pedro]

Thanks. This would do for me. Having never shot a 5D2, from what I see in posts 25k especially in BW looks pretty well. As I come from the filmdays and do like Robert Frank's photography it would seem like the perfect camera to me at a "moderate" price point for amateurs: 3 to 3.5 k.

 

[AprilForever]

I'm an ISO'er.
50% of what I do is shooting Rock bands. The headline acts are always well illuminated and never an issue, but the guys who come on first sometimes get nothing more than 1 red and 1 green 100W "floodlight", and an occasional other 100W light rotating through colours .
To be able to capture them with minimal blur and noise would make them, and me, very happy
The other 50% is Birds in Flight. Again, shooting between April and September is no problem, but for the winter....Unless it's one of those rare bright blue days, or it's snowed, I leave my camera at home

What's wrong with shooting birds when its overcast? Some of my best pictures are taken in the winter on an overcast day...

 

[Mt Spokane Photography]

While opinion and wishes are nice, and Yes, I'd like a higher MP image, I've seen nothing in the long rambling article to convince me that with current technology, cameras and lenses, that we will continue to gain in resolution with smaller photosites.

In fact, the person who invented CMOS, Eric Fossum, is now a Yale professor, and has a list of inventions, papers, and awards flatly states that there is a limit to the size of photosites that will result in higher resolution due to diffraction from the lenses we currently use.

http://en.wikipedia.org/wiki/Eric_Fossum

<iframe width="560" height="315" src="http://www.youtube.com/embed/JkBh71zZKrM" frameborder="0" allowfullscreen></iframe>


http://academic.research.microsoft.com/author/678550?form=BA

Who would you believe??

 

[thepancakeman]

Who would you believe??

Rarely the naysayers. History is replete with philosophers and scientiest and inventors saying "it can't be done" only to have someone else do it.

 

[Loswr]

Who would you believe??

The people on this forum, obviously, like the guy who said that the lens doesn't factor into the effect of diffraction.... Why would the guy who invented CMOS sensors know anything about them and their limitations?

 

[thatcherk1]

Someone please tell me if I'm missing something. I've always understood pixel density and noise to be related in this way:

Below is a non-real world example.

If I have a 5MP FF sensor, I will have x amount of noise at ISO100. If i have that same size sensor, but instead it's 20MP, I now have 4x the pixels, and they are 1/4 the size of those in the 5MP sensor. Each pixel receives 1/4 the amount of light that the 5MP sensor pixels did because their surface area is 1/4 the size. So to get the same brightness, the sensor needs 4x the amplification to read the same brightness measurement of ISO100. This makes the 20MP image 4x noisier than the 5MP image when viewing the two side by side at 1:1pixel view on a monitor, or lightjet printer. The big difference is that the 20MP image is 200% bigger at 1:1 than the 5MP image. Now if you want to convert the 20MP image to be the same resolution as the 5MP image, you can use Photoshop and it's smart algorithms to take every group of 4 pixels and average them to make 1 pixel. Now, if you take 4 high-noise pixel samples and create one new one by averaging them, you will get a pixel that is 4x the accuracy of any single one of those pixels, thus rendering the noise to be 1/4 of the pronouncement. Noise by definition is a variation in pixel brightness (per channel) that is introduced into an image due to imperfect electronic capture. So averaging 4 varying-brightness pixels into 1 pixel will dramatically increase the accuracy of that one pixel.
In doing this down-sampling from a 20MP to 5MP image, you essentially get 1/4 the noise, and 50% the resolution, thus matching that original 5MP image that was captured with 1/4 the amplification and 50% the resolution of that of the 20MP captured image.

If the above example is true, then I can have a 20MP camera that shoots an acceptable noise level at ISO100, I can blow it up nice and big, and look at it nice and close. Or if I'm in a low light situation shooting ISO3200 and my 20MP gives me more than the acceptable amount of noise, I can down-res the image to 50% size in order to kill much of the noise. I can do this in photoshop/lightroom, or in-camera with mRAW or sRAW, thus also yielding me smaller file sizes.

There are several more factors in discussing resolution and MP usability: lens sharpness, moire patterns, OLP filter, viewing distance at different print sizes, camera processing speed, file size, etc.

But the above is a way to hopefully look only at the relationship between Signal-to-noise ratio vs. pixel density.

The two variables that might cause my logic and calculations to be flawed are A. the rate of noise introduction when doubling voltage to a sensor. If it's 1:1 ratio of noise intensity to voltage increase, then my calculations are correct. And B. If when you make a pixel 1/4 the size, does it receive less than 1/4 the light. If anyone has any input on this please let me know.

A third variable is, if you average 4 pixels that are 1/4 the size and 4x the noise of another single pixel, do you actually end up getting an even more accurate pixel than the one larger pixel due to some kind of law of averages?

note: I didn't use the term photo-site at all. I just used pixel as a universal term to refer to photo-sites and digital pixels.

Also, the above example is based on quadrupling the MP and pixel density because it's easier to understand than just doubling. But the principle still applies to any rate of multiplication of pixel density.

I can show photo examples of my principle stated above if anyone would like a better explanation.

Please feel free to comment on my thoughts above and where I might have gone wrong. But don't be a jerk. I personally shoot for a newspaper where ISO is more important than resolution, but my personal photography/art is high res, low ISO landscape and portraiture. So I have no specific desire for one type of camera or another.

 

[jrista]

what do u think of this statement:

http://www.kenrockwell.com/tech/mpmyth.htm

I think Rockwell is a self-proclaimed twit, and I take his articles in that spirit.

 

[thepancakeman]

what do u think of this statement:

http://www.kenrockwell.com/tech/mpmyth.htm

He says "Today, even the cheapest cameras have at least 5 or 6 MP, which enough for any size print. How? Simple: when you print three-feet (1m) wide, you stand further back. Print a billboard, and you stand 100 feet back."

And how, exactly does he know where I stand? 8)

 

[thatcherk1]

what do u think of this statement:

http://www.kenrockwell.com/tech/mpmyth.htm

I think Rockwell is a self-proclaimed twit, and I take his articles in that spirit.

He certainly has some decent articles that I've learned from. But you have to weed out his personal biases with actual universal facts of information. I wouldn't go as far as "twit", but certainly biased and "sure of himself".

 

[jrista]

I'm one of those "MPers", that said I would be satisfied nif the 5dmkIII was 25MP. Although it'd be better to combine the two - high MP and high, clean ISO without losing image quality. But hey, we can't have it both ways (maybe in the next 10-15 years we can). Just my 2 cents

I believe it would be EASY to have both 25mp and clean ISO. Noise is a combination of factors...some we cannot control, some we can control to a degree, and some we have a lot of control over. The noise floor of recent Sony sensors is around 12% that of Canon sensors prior to the 1DX. Mainstream Canon sensors have a funky way of boosting ISO, and in many cases they perform analog gain amplification when electronic read noise is quite high, amplifying all that noise right along with the rest of the image. Reordering circuits, cooling circuits down, etc. can all have a positive impact on noise levels, even at higher resolutions than we have today. I believe the technology already exists in the 1DX sensor, so a lower-FPS, low-ISO sensor that doesn't need ISO52100 should be able to see considerable gains in ISO levels up to ISO6400 or so.

 

[briansquibb]

Lets go a very large, low density, low noise sensors which would be cheap to manufacture. Maybe 6x4 size

Ah.... Canon are rumoured to be going that way

 

[Mt Spokane Photography]

Lets go a very large, low density, low noise sensors which would be cheap to manufacture. Maybe 6x4 size

Ah.... Canon are rumoured to be going that way

The cost would be enourmous. The size is the biggest factor in cost by far. And, are you ready to buy $15K lenses?

 

[AprilForever]

I'm one of those "MPers", that said I would be satisfied nif the 5dmkIII was 25MP. Although it'd be better to combine the two - high MP and high, clean ISO without losing image quality. But hey, we can't have it both ways (maybe in the next 10-15 years we can). Just my 2 cents

I believe it would be EASY to have both 25mp and clean ISO. Noise is a combination of factors...some we cannot control, some we can control to a degree, and some we have a lot of control over. The noise floor of recent Sony sensors is around 12% that of Canon sensors prior to the 1DX. Mainstream Canon sensors have a funky way of boosting ISO, and in many cases they perform analog gain amplification when electronic read noise is quite high, amplifying all that noise right along with the rest of the image. Reordering circuits, cooling circuits down, etc. can all have a positive impact on noise levels, even at higher resolutions than we have today. I believe the technology already exists in the 1DX sensor, so a lower-FPS, low-ISO sensor that doesn't need ISO52100 should be able to see considerable gains in ISO levels up to ISO6400 or so.

I'm hoping for ISO 51,200 on my 7D mkII! With 24MP. And I am sure it will work great and look great; moreover, I am certain that people will continue to complain about the noise levels in the blue skies at ISO 400...

 

[jrista]

Someone please tell me if I'm missing something. I've always understood pixel density and noise to be related in this way:

Pixel density is a factor of image noise. There are many things that affect image noise:

* Photon shot noise (randomness of light particles)
* Thermal noise (random noise generated by heat from the sensor and other electronic components)
* Electronic noise (noise generated by electronic circuits themselves)

Photon shot noise is something we can't control...its a matter of physics and the nature of light. The other types of noise we can control. Thermal noise is pretty easy...we can use passive or active cooling technology to draw heat away.

With higher density sensors, there are more electronic circuits packed into the same area, so that increases electronic noise. That kind of noise can be controlled to a certain degree by changing how we fabricate sensors, how we design column/row activate and readout logic, in what order we amplify the signal, convert it to digital, etc. If it was truly the case that the more pixel dense a sensor was the noisier it was, sensors would have become unusably noisy LONG ago. At 100% on a per-pixel basis, the 7D appears to be fairly noisy compared to say a 50D. In reality, the amount of additional noise in the 7D is only marginally greater than the 50D at ISO100 (a matter of a fraction of a percent), not to mention that noise tends to be more even in the 7D than even the 5D II. And the 7D uses older technology to handle readout, amplification, and ADC.

Canon demonstrated a couple of radical new prototype sensor designs this year, one with an unbelievable amount of low-light sensitivity, and one with 120mp packed into an APS-H sensor, with manageable noise and a reasonable readout rate. I think tech from both those prototypes have made it into the 1DX 18mp sensor. Its obvious that they have managed to greatly improve the levels of noise in the sensor, as it supports 3 stops more light sensitivity at the same noise levels as prior ISO6400 sensors.

The story with noise is not quite as simple as "more density means more noise". There are a multitude of factors that contribute to noise, and many of them we can control. The 1DX 18mp sensor is considerably better than the 16mp 1D IV sensor, even though its higher density. I think that lends some strong evidence in support of the idea that we can get both better ISO/lower noise and more MP at the same time.

 

[KeithR]

I think that lends some strong evidence in support of the idea that we can get both better ISO/lower noise and more MP at the same time.

And we see the exact same trend everywhere:

7D better than 40D better than 30D better than 10D;
Nikon D7000 better than D300 (much!) better than D200;
Pentax K-5 better than K-x;
5D Mk II better than 5D;
1D Mk IV better than Mk III better than Mk II better than 1D.

And so it goes, always in the same direction - newer cameras, more pixels, better noise performance.

Yet the Flat Earthers still argue that if we want better noise capability we need fewer pixels. Never mind the evidence, they just shut their eyes, stick their fingers in their ears and spout the same old crap regardless...

 

[Mt Spokane Photography]

I think that lends some strong evidence in support of the idea that we can get both better ISO/lower noise and more MP at the same time.

And we see the exact same trend everywhere:

7D better than 40D better than 30D better than 10D;
Nikon D7000 better than D300 better than D200;
Pentax K-5 better than K-X;
5D Mk II better than 5D;
1D Mk IV better than Mk III better than Mk II better than 1D.

And so it goes, always in the same direction - newer cameras, more pixels, better noise performance.

Yet the Flat Earthers still argue that if we want better noise capabilities we need fewer pixels...

I've had Digital cameras for many years, since the 1990's, and noise does indeed go down with each new generation. However, it only drops a little. There is little difference from generation to generation, because, in addition to new technology, the increase in MP off sets what might otherwise be a larger gain.

Balancing the noise and light sensitivity against gains in resolution by adding more MP is what we are discussing. As photosite sizes decrease, we have new technology to off set some of the detrimental effects.

However, none of the noise reduction technology ofsets the degradation due to lens diffraction. Fortunately, its a gradual degredation, not a sharp cutoff, so the complex relationships between these things are only determined by the sensor and camera manufacturers who build prototypes for that purpose.

There are two sides, the sales side knows more pixels sell, while the engineering side knows that there is a point of diminishing returns to adding more MP. No one knows exactly where that point is, but around 4 microns has been mentioned as a reasonable figure by the Engineering experts. That number may go lower.

Its not a matter of designing a better lens, all lenses have diffraction, its just that with smaller photosites, we see the effects better until we reach a point where it actually reduces resolution.

You can see this very well by viewing some of the test figures from lens testers. The resolution of a lens appears to drop with smaller and smaller apertures. What causes this is diffraction.

As bodies get smaller photosites, the point at which resolution decreases comes at a larger aperture. This means that there is a point where smaller photosites (More MP) no longer increases resolution. It does not get worse, you just don't get improvement proportional to the MP increase.


This is engineering, not Flat Earther thinking, its been determined over and over again by facts.

Calling people names is only the mark of childish thinking.

 

[jrista]

While opinion and wishes are nice, and Yes, I'd like a higher MP image, I've seen nothing in the long rambling article to convince me that with current technology, cameras and lenses, that we will continue to gain in resolution with smaller photosites.

In fact, the person who invented CMOS, Eric Fossum, is now a Yale professor, and has a list of inventions, papers, and awards flatly states that there is a limit to the size of photosites that will result in higher resolution due to diffraction from the lenses we currently use.

http://en.wikipedia.org/wiki/Eric_Fossum

<iframe width="560" height="315" src="http://www.youtube.com/embed/JkBh71zZKrM" frameborder="0" allowfullscreen></iframe>


http://academic.research.microsoft.com/author/678550?form=BA

Who would you believe??

Sure, there are physical limits, no question about that. However, the highest density DSLR sensors these days resolve well below the theoretical maximum at Rayleigh (MTF 9%) for a perfect lens. A perfect diffraction-limited f/2.8 can resolve about 523 lp/mm at 9%, however the 7D's 18mp nyquist rate is 116 lp/mm, and the Sony A77's 24mp nyquist rate is 128 lp/mm. Granted, modern lenses are not "perfect", but they are damn good. A perfect lens @ 9% at f/11 resolves 135lp/mm, still above the rate of the Sony 24mp APS-C sensor. That tells me there is room to grow, so long as lenses keep approaching diffraction-limited perfection.

 

[jrista]

I guess I should clarify my position here, as some comments seem to assume a different position. I am not saying every camera should always have more MP. Thats not the case at all. What I am saying is that there shouldn't be a single focus from camera manufacturers. Right now, the rumors seem to indicate that Canon is REDUCING MP on most of their sensors in favor of RADICALLY improving ISO. Its GREAT that Canon is listening to many of their customers by doing so with the 1DX, and for some of the things I do, a lower MP, better ISO, lower noise camera will serve me perfectly.

However, I do not think its good for Canon to solely and unequivocally cater to ONE vocal group...the ISOers, at the cost of the needs of other groups, like the MPers. For that matter, I don't think its good if they are focusing intensely on ISO and low noise if it costs dynamic range, as many people need better dynamic range at low ISO far more than they need high MP res or low-light performance.

My argument here is that its not possible to create a one-size-fits-all sensor, and camera manufacturers shouldn't try. The ISOers got their wish with the 1DX and its fantastic lower-res, high-ISO sensor. However, every time someone opens their mouth to ask for a high MP 5D III, the ISOers come out of the woodwork to beat them down with anecdotal evidence that more MP offers no benefit, and that the only thing that could ever make sense these days is a lower-res 5D III with the same ISO/noise characteristics as the 1DX (oh, yeah, and 7D AF...which is again just as naive.)

 

[eaw213]

7D better than 40D better than 30D better than 10D;
Nikon D7000 better than D300 (much!) better than D200;
Pentax K-5 better than K-x;
5D Mk II better than 5D;
1D Mk IV better than Mk III better than Mk II better than 1D.

And so it goes, always in the same direction - newer cameras, more pixels, better noise performance.

Yet the Flat Earthers still argue that if we want better noise capability we need fewer pixels. Never mind the evidence, they just shut their eyes, stick their fingers in their ears and spout the same old crap regardless...

I wouldn't quite say that's what people are arguing for. At least, it's not what I'm arguing for. Think of it as points you get to apply to upgrades on a new camera. Canon gives you three points, and you can upgrade technical features, ISO quality, or MP. You say that you'd apply one point to each category, and get a camera better at everything! More features, more MP, and still better ISO. Okay, fine. If it were me though, I'd apply one point to new snazzy features, two points to ISO, and no points to MP increase. I don't think it's quite fair to claim people would apply one point to features, one point to ISO, and non to MP just to spite you. In other words, I think what people argue for is an even bigger jump in ISO performance at the expense of more MP.

Consider the following statement:

1. "If you were to compare a 1DX as designed now with 18MP with a 1DX that has 36MP, I think the ISO would be better in the 18MP version."

I think the statement above is true, but it's important to note that it says nothing about what I think of the 36MP 1DX compared to the prior cameras. If you're curious, I would also add this statement

2. "I think even a 36MP 1DX would still have better ISO than prior cameras."

I think the two statements are perfectly valid and can coexist as well. I think the main difference comes down to whether we are talking about "better" ISO or "even better" ISO. You'll get better ISO either way, but if you want to really start focusing on ISO, then not having big or any MP jump might be one possible path to go down. It's all a balancing act.