What about 36MP DSLRs?

[ErikKaffehr]

Hi!

I just posted an article predicting image quality on the coming generation of 36 MP cameras, based on 16 MP APS-C.

http://echophoto.dnsalias.net/ekr/index.php/photoarticles/60-what-about-36mp-dslrs

Best regards
Erik

[BernardLanguillier]

Thanks, it does for sure show that accuracy of focus will be critical to tap in the resolution potential.

It will also be interesting to see if some if the rumors related to a non AA filter version of the D800 are true or not.
 
Finally, lens quality will become very critical, especially color aberation in corners.

Cheers,
Bernard

[jeremypayne]

Great job with this one ... very interesting post.  One issue - when you click on the teapot, you get an error.

[ErikKaffehr]

Hi Bernard,

Just a few observations. The Sony Alpha SLT 55 I used has a weak AA filter and I see quite a lot of aliasing artifacts. So I don't know if 4 micron pixels really grant removal of AA-filtering. But the AA-filter is not exactly cheap and some people don't want it. Regarding lens quality it will be critical, but it is already critical on 16 MP APS-C.

Lateral chromatic aberration can be easily handled in raw conversion on unshifted lenses, with TS it would be worse.

On point I'd make that this images were shot with focus bracketing. I used a Sony 50/1.4 it's decently sharp when stopped down, but at f/1.4 it has a lot of axial chromatic aberration and probably some focus shift. This is less a problem with AF but focus shift may be a problem with "live view" unless you focus stopped down.

Best regards
Erik

Thanks, it does for sure show that accuracy of focus will be critical to tap in the resolution potential.

It will also be interesting to see if some if the rumors related to a non AA filter version of the D800 are true or not.
 
Finally, lens quality will become very critical, especially color aberation in corners.

Cheers,
Bernard

[torger]

I wonder what type of quality one can expect in fullframe lens corners at say f/8-f/11 of good quality primes. With APS-C you only see the center portion.

There's always quality drop towards the lens corners, some suggest that this quality drop is so large even on good lenses that 36 MP is clearly over the top. I think it will be good enough, but I don't really know how to estimate that type of performance without actually having a 36 MP fullframe sensor.

[ErikKaffehr]

Hi,

Yes that may be a concern. My take is that high quality lenses on the long side are normally excellent across the image. Wide angles are problematic. Some extreme wide angles are very good, Nikon 14-24/2.8 and Zeiss 21/2.8 come to mind, but most are not so good in the corners.

In general, there is only one plane of focus and that is where you can achieve maximum sharpness. So for landscape and architecture with wide angles that is probably an issue. The solution is:

- Get good lenses
- Stop down to best aperture

On the other hand, going from 24 MP to 36 MP is not a dramatic step, a 22% increase in resolution. The major benefit may be less jaggies, less aliasing and that the image sharpens and scales better. It's better that the sensor outresolves the lens than the other way around, you get less aliasing.

Best regards
Erik

I wonder what type of quality one can expect in fullframe lens corners at say f/8-f/11 of good quality primes. With APS-C you only see the center portion.

There's always quality drop towards the lens corners, some suggest that this quality drop is so large even on good lenses that 36 MP is clearly over the top. I think it will be good enough, but I don't really know how to estimate that type of performance without actually having a 36 MP fullframe sensor.

[torger]

It's better that the sensor outresolves the lens than the other way around, you get less aliasing.

Couldn't agree more :-). I do hope these sensors materialize. With high res sensors I think we will see more development in the deconvolution field too, to be able to reverse lens and diffraction blur to some extent.

[ErikKaffehr]

Hi,

Here is an interesting posting by Diglloyd: http://www.diglloyd.com/blog/2011/20111121_1-Megapixels.html

Small comment. Going from 24 MP to 36 MP is a minor step. It's a 22% increase in resolution. No dramatic effects expected.

Best regards
Erik

[BernardLanguillier]

Small comment. Going from 24 MP to 36 MP is a minor step. It's a 22% increase in resolution. No dramatic effects expected.

That's true. But it seems to be seen as being truer when the jump is done between 2 DSLRs as opposed to being done from a DSLR to a so called MF camera.  

Realistically though, it is also going to be true that the jump from a hypothetical 36 MP D800 to a 50 MP H4D50 will be even more insignificant. Insignificant for demanding very large print applications and even more insignificant for normal size prints... like A2.

That is based on the assumption of a perfect focus, perfect technique,... that is typically not achieved in many images because many people just cannot afford to spend the increasingly longer time required to tap in those types of resolutions in a single frame.

The painful truth remains that stitching is the only significant way forward in resolution for those applications needing real detail that are compatible with the technique.

Of course, 36 can also be seen as being 3x12 which could have different implications. 


Cheers,
Bernard

[ErikKaffehr]

Hi,

I mostly mean that increasing resolution by 22% doesn't subjects lenses to enormous stress.

Regarding MF, it's also much about lenses. Best lenses on DSLRs may match the lesser lenses on MFDBs, but in general I would suggest that MF has an advantage of size.

3x36 is 108, BTW. I also know my math ;-)

Best regards
Erik

That's true. But it seems to be seen as being truer when the jump is done between 2 DSLRs as opposed to being done from a DSLR to a so called MF camera.  

Realistically though, it is also going to be true that the jump from a hypothetical 36 MP D800 to a 50 MP H4D50 will be even more insignificant. Insignificant for demanding very large print applications and even more insignificant for normal size prints... like A2.

That is based on the assumption of a perfect focus, perfect technique,... that is typically not achieved in many images because many people just cannot afford to spend the increasingly longer time required to tap in those types of resolutions in a single frame.

The painful truth remains that stitching is the only significant way forward in resolution for those applications needing real detail that are compatible with the technique.

Of course, 36 can also be seen as being 3x12 which could have different implications. 


Cheers,
Bernard

[deejjjaaaa]

The painful truth remains that stitching is the only significant way forward in resolution for those applications needing real detail that are compatible with the technique.

superresolution too to some extent and imperfections are helpful for it (provide necessary shifts).

[buldoozer]

High.,Iam new on this forum and i come from the Netherlands so forgive me my bad englisch.I think at normal print sises you wont be able too see the differance between 36MP and 18MP FF sensors because they have both the same sensor sise.Only on very very big enlargements you whil see more detail in the 36 version.I also think that lens resolution whil not be the limiting factor,take a look at Clarkvision.com , he explanes it oll.The test that now becomes interresting is between FF 24x36 mm digital sensor and FF digital medium format sensor.Both have no digital grain orr noise anymore at 100 ISO.I think at normal print sises you wont see the differance anymore because the limiting factor between 24x36 and 6x4.5 medium format analoog always was the grain and not the lens .the sise of the film only becames importend whit very large prints.This limiting grain factor has gone now.So just like the film sise the sensor sise becomes only importent at very very large prints,like 60x90 inch orr more.

[ErikKaffehr]

Hi,

The simplest way to see it is that doubling resolution will allow you to go to the next A-size. So if 18 MP is good for A2 you would be able to A1 with 36 MP. This would be the case if optics were aberration free, and medium aperture were used so diffraction would not set in.

Viewing distance is much involved in this. You could make very large print as long as viewing distance is increased proportionally to print size. This is not really the case, I guess that we see larger prints at relatively close distance.

Increasing sensor size has two advantages:

1) A larger sensor sees more photons. Most noise in digital images comes from the natural variation of incoming photons, so noise will reduce with a larger sensor, all other factors kept constant.

2) A larger sensor will be used with a longer focal length. So the image of any structure will be larger on the sensor. The contrast the lens can transfer increases with image size, so fine detail contrast will be better with a larger sensor, again, all other factors kept constant.

Best regards
Erik

High.,Iam new on this forum and i come from the Netherlands so forgive me my bad englisch.I think at normal print sises you wont be able too see the differance between 36MP and 18MP FF sensors because they have both the same sensor sise.Only on very very big enlargements you whil see more detail in the 36 version.I also think that lens resolution whil not be the limiting factor,take a look at Clarkvision.com , he explanes it oll.The test that now becomes interresting is between FF 24x36 mm digital sensor and FF digital medium format sensor.Both have no digital grain orr noise anymore at 100 ISO.I think at normal print sises you wont see the differance anymore because the limiting factor between 24x36 and 6x4.5 medium format analoog always was the grain and not the lens .the sise of the film only becames importend whit very large prints.This limiting grain factor has gone now.So just like the film sise the sensor sise becomes only importent at very very large prints,like 60x90 inch orr more.

[torger]

Concerning print sizes viewing distances etc it depends very much on what context the image is shown and what the image content is.

A portrait showing a face does not need very high resolution, there's little interest in the small details (pores in the skin), and you want to see the whole face at once. So 10 megapixels or so is probably more than enough for any print size.

Another extreme would be a wide panorama print (say 3:1 format) of a landscape view, mountains in the distance, a village in a valley, roads with cars, trees etc, a large print framed on the wall at eye height. Then it is natural to walk up close to look at small details, and you can look at the image one part at a time.

You don't need large views from mountain tops though, pretty much any landscape picture can apart from being viewed as a whole also be appreciated up close, let the eye wander around in the picture. A high res print is like a hifi audio recording, the artistic content (picture / music) does not get better, but the quality of the medium still adds value.

[ziocan]

Nice experiment!
It reminds me of some comparative tests i did between the a900 and the Phase P30 back which is about 32mp without aa filter.
I would say that the difference in detail is very similar. Tangible, not compelling, but nevertheless a plus.
On A3 size print, which is about a magazine spread, at an arm lenght distance, it showed some difference in detail to my eyes. of course was barely noticeable.
If that will be a must have, it will depend on the photographer.
If Sony will offer that sub 4000$, I will not pass it.

The petals may not show much of a difference, but the leaves do. IMO

[ErikKaffehr]

Hi,

Yes, I observed that too. It may have to do with lens having chromatic aberration affecting the reds or simply on red light having longer wavelength and thus being more affected by diffraction. Focus should be essentially the same as focus bracket was used at one centimeter intervals at about 1.5 m for both images. Best resolution on test target has been chosen.

I doubt very much if the two could be told apart in A2 prints, but I have not yet made the actual experiment.

Best regards
Erik

Nice experiment!

The petals may not show much of a difference, but the leaves do. IMO

[ErikKaffehr]

I agree, absolutely!

Erik

Concerning print sizes viewing distances etc it depends very much on what context the image is shown and what the image content is.

A portrait showing a face does not need very high resolution, there's little interest in the small details (pores in the skin), and you want to see the whole face at once. So 10 megapixels or so is probably more than enough for any print size.

Another extreme would be a wide panorama print (say 3:1 format) of a landscape view, mountains in the distance, a village in a valley, roads with cars, trees etc, a large print framed on the wall at eye height. Then it is natural to walk up close to look at small details, and you can look at the image one part at a time.

You don't need large views from mountain tops though, pretty much any landscape picture can apart from being viewed as a whole also be appreciated up close, let the eye wander around in the picture. A high res print is like a hifi audio recording, the artistic content (picture / music) does not get better, but the quality of the medium still adds value.

[GeraldB]

Would down sampling in post processing from 36mp to say 24 mp improve the noise? i.e. would it be possible to shoot at a higher ISO for the same noise? If this is true how does one calculate by how much i.e. image shot at ISO 3200/36mp would be similar in noise to 24mp/??mp.

[BernardLanguillier]

Would down sampling in post processing from 36mp to say 24 mp improve the noise? i.e. would it be possible to shoot at a higher ISO for the same noise? If this is true how does one calculate by how much i.e. image shot at ISO 3200/36mp would be similar in noise to 24mp/??mp.

Yes, it is going to decrease the noise, and therefore increase DR.

How much should be possible to compute, but I don't have a clear answer.

Regards,
Bernard

[ErikKaffehr]

Hi,

It depends. Shot noise is essentially independent of pixel size for a given size of sensor and Quantum Efficiency. Think of shot noise as noise lights and mid tones.

Regarding read noise (darkest part of the image) the math is different. Increasing pixel size will reduce noise more than just resampling to lower density. Modern sensors may have much lower readout noise than older sensors. Take for instance the Nikon D7000 which has a very low noise sensor.

Many cameras have pretty high read noise and those can be helped by preamplifiers for high ISO. Canon cameras are like that. Read noise is low at base ISO but is reduced with increasing ISO, so shadow detail is not much affected by increasing ISO from say 100 to 400. Shot noise will increase with increasing exposure.

So the answer to the question depends much on in camera electronics.

The enclosed diagrams demonstrate the difference. DR is based on maximum signal / read noise and corresponds to the darkest part of the image. Tone depth is essentially shot noise.

Both decrease with increasing ISO, but hardware amplification helps Canon at ISO up  to around 800, Nikon does not need it. Would the Nikon D7000 be a full frame camera it would have 36 MP. The Canon 7D sensor scaled to full frame would yield 46 MP.


Best regards
Erik

Yes, it is going to decrease the noise, and therefore increase DR.

How much should be possible to compute, but I don't have a clear answer.

Regards,
Bernard