Has Canon fallen hopelessly behind? (cross-posted)

[BJL]

in my opinion big pixels are better,  it will be good to compare  RAWs at f/16
and who needs more than 22mp  ? 0.1% of photographers ?

As I indicated in another reply, comparing different formats at the same very high f-stop is pointless: the principal reason for using such a small aperture is tomget lots of DOF, and for that, what needs f/16 in 645 format only needs about f/8 in 35mm format.

As to the question of who needs more than 22MP, I agree that it is a small minority of all photographers, and does not include me. But amongst the small proportion who do have a use for more than 22MP, many might prefer to get it in at the price of a Nikon D800 system rather than with a far more expensive medium format system, especially when the Nikon option offers far better low light handling, less expensive lenses, and probably even better dynamic range, due to the vast technolical gap that has opened in recent years between CMOS sensors and the obsolescent CCD technology that DMF has been stuck with so far. Many of the traditional advantages of bigger pixels (assuming equal technology) are off tue table so long as the sensors with those bigger pixels have vastly more read noise.

[ErikKaffehr]

Hi,

No, the T stop doesn't really matter for noise as long as exposure is the same. A lower T-stop allows for shorter exposure times or using lower ISO.

Best regards
Erik

How about the role of the lens in concentrating light?

I would think that the size of the sensor is in fact irrelevant. What matters is the true T stop of the lens, is it not?

Cheers,
Bernard

[BernardLanguillier]

Hi,

No, the T stop doesn't really matter for noise as long as exposure is the same. A lower T-stop allows for shorter exposure times or using lower ISO.

The amount of photons is controlled by the surface of the front glass of the lens, correct?

What happens between that moment and the landing on the sensor?

Cheers,
Bernard

[BJL]

Besides bigger pixels have less noise, better high ISO.

Another truism that gets repeated without apparent reference to faactsmor counter-arguments. Tests ofmtue D800 show that, when a lower resolution is sufficient so that 22MP is enough, downsampling to that lower pixel count essentially equalizes the npise levels and high ISO performance. Combined with the option to get higher resolution when noise levels will still be acceptable, it is hard to see noise as an argument against higher resolution.

For landscape photography sharp, small sensors are a problem as stopping down brings the effective resolution to around 5 MPix only (f:16-22 with FF sensor). With larger sensor with same MPix you can stop down more, which compensates most of the lost DOF ...

I very much doubt that most use of a camera like the D800 will require stoppoing down to f/16-22, but when it does, you are right that the extra stopping down in a larger format can give the sam balance of diffraction limits on resolution againsts DOF. There is another dogma that larger formats are worse for getting great DOF, based on the recurring error of comparing at equal f-stop, but in fact the only issue for a larger format is that the higher f-stops needed require some combination of higher Exposure Index (confusingly called ”ISO") and longer exposure times. With similar sensor technologies, the larger format can get the same shutter speed by using a higher EI and still get comparable noise levels due to the "large pixel advantage". However, this is far from true in comparisons between recent 35mm format CMOS sensors and medium format CCDs, so that for now, MF is stuck with needing longer exposure times (or accepting worse noise) to get equally large DOF.

[BJL]

Erik and Bernard,

As far as total photons counted from the subject, which as we all seem to agree is what determines the shot noise, regardless of how many sensels we gather them in, or the size of those sensels:
1. With full exposure at low exposure index, making optimal use of the full well capacity of the sensels, T-stop is irrelevant. A higher T-stop, due for example to a zoom lens that loses more light to internal reflections, simply requires a longer exposure time to fill the highlight sensels.
2. When requirements of adequately high shutter speed force the use of a higher EI, so undefilling the sensels, then T-stop matters. In fact, for cross-format comparisons, a nice single number measure would be an "adjusted effective aperture diameter", being focal length divided by T-stop. Why? For equal exposure time of the same scene with same lighting, the photon count reaching the sensor is proportional to the square of this. (To get a bit more mathematical, pi/4 times the square of this adjusted effective aperture diameter is an adjusted effective aperture area, and this is the "area" that measures which fraction of the photons from the subject are caught by the lens and delivered to the sensor.)

For most purposes, the above could just be done with f-stops instead of the cinematographer's T-stops, and then the quantities I am talking about are the "effective aperture diameter" and "effective aperture area", standard quantities in scientific studies of photographic optics.

To Bernard in particular: these measures of "effective aperture area" are the relevant measures of "the surface of the front glass of the lens" that you referred to.

[BernardLanguillier]

Erik and Bernard,

As far as total photons counted from the subject, which as we all seem to agree is what determines the shot noise, regardless of how many sensels we gather them in, or the size of those sensels:
1. With full exposure at low exposure index, making optimal use of the full well capacity of the sensels, T-stop is irrelevant. A higher T-stop, due for example to a zoom lens that loses more light to internal reflections, simply requires a longer exposure time to fill the highlight sensels.
2. When requirements of adequately high shutter speed force the use of a higher EI, so undefilling the sensels, then T-stop matters. In fact, for cross-format comparisons, a nice single number measure would be an "adjusted effective aperture diameter", being focal length divided by T-stop. Why? For equal exposure time of the same scene with same lighting, the photon count reaching the sensor is proportional to the square of this. (To get a bit more mathematical, pi/4 times the square of this adjusted effective aperture diameter is an adjusted effective aperture area, and this is the "area" that measures which fraction of the photons from the subject are caught by the lens and delivered to the sensor.)

For most purposes, the above could just be done with f-stops instead of the cinematographer's T-stops, and then the quantities I am talking about are the "effective aperture diameter" and "effective aperture area", standard quantities in scientific studies of photographic optics.

To Bernard in particular: these measures of "effective aperture area" are the relevant measures of "the surface of the front glass of the lens" that you referred to.

Thanks for the explanation.

So in essence, when a lens is used, sensor size is not a relevant factor to assess the amount of light reaching the sensor, and therefore not a relevant factor to assess shot noise, correct?

Cheers,
Bernard

[Petrus]

How many times does this myth have to be debunked?!

What I naturally meant was that diffraction problem gets worse in the sense that you are not getting the promised resolution if you turn the aperture too small. Of course the final resolution is not worse than with a less dense sensor, but it is not better either.

What comes to small versus large sensor with the same MP figure and stopped down to an aperture for the same DOF, what is the exact truth in this matter? Of course, again, we have to use smaller stop with the larger sensor camera, but do the DOF and diffraction walk hand in hand, or does one system pull ahead?

One myth which creeps up is the need for better focus with sharper systems. That is not true either, as focus is critical only with bigger than normal enlargements. With lesser systems you need to focus just as well, lack of resolution just does not show you missed!

[BJL]

Thanks for the explanation.

So in essence, when a lens is used, sensor size is not a relevant factor to assess the amount of light reaching the sensor, and therefore not a relevant factor to assess shot noise, correct?

Bernard, [Edit: Bob Fisher take note too!]

    You give me the opportunity to say something nice about larger formats in general and MF in particular, in case I have come accross as too cynical on that front in recent posts:

When the sensor can be given "full exposure", a larger sensor typically has a greater total well capacity (electons per sensel times sensels per sensor) and so can count more electons and get a better ratio of signal to shot noise, and in all but extremely deep shadows, this should win out over read noise and give a better SNR. The only trade-off for tue laeger format is either a longer exposure time (if for example you equalize DOF by using a higher f-stop in the larger format, or the larger format lenses limit you to a higher minimum f-stop), or less DOF (if for example you use equal f-stop).

And even when the tecnhological gap between modern CMOS sensors and CCDs is taken into account, I am fairly sure that CCDs can count roughly as many electons per unit area, and so win on maximum total counts of electons and thus of photons. It even more clear that the best 35mm formats sensors win over smaller formats for shot noise control at minimum EI.

[RFPhotography]

Thanks for the explanation.

So in essence, when a lens is used, sensor size is not a relevant factor to assess the amount of light reaching the sensor, and therefore not a relevant factor to assess shot noise, correct?

Cheers,
Bernard

Bernard, I'm not sure I understand your question in relation to Erik's response.  Maybe I'm reading Erik's response incorrectly.  Sensor size may not be a relevant indicator of light reaching the sensor through the lens (that, of course makes sense) but it is a relevant indicator of how many photons are captured which is what determines shot noise.  So is it right to say that sensor size is irrelevant?  I'd come at it from a different direction; all else being equal it is sensor size that determines shot noise. 

[BJL]

What I naturally meant was that diffraction problem gets worse in the sense that you are not getting the promised resolution if you turn the aperture too small. Of course the final resolution is not worse than with a less dense sensor, but it is not better either.

...

One myth which creeps up is the need for better focus with sharper systems. That is not true either, as focus is critical only with bigger than normal enlargements. With lesser systems you need to focus just as well, lack of resolution just does not show you missed!

Yes, I see what you mean now: making full use of extra resolution requires extra effort and restrictions (more careful focusing, better control of camera motion and freezing subject motion, f-stops that avoid excessive diffraction, etc.), but on the other hand, if you operate the same as with a lower resolution sensor including displaying at the same size, nothing gets worse; you have at worst squanded some or all of the potential advantage.

What comes to small versus large sensor with the same MP figure and stopped down to an aperture for the same DOF, what is the exact truth in this matter? Of course, again, we have to use smaller stop with the larger sensor camera, but do the DOF and diffraction walk hand in hand, or does one system pull ahead?

Ignoring factors like lens abberations, and the quirks at very close focusing range, it is a tie: adjusting focal length in proportion to sensor size to get the same FOV, and then adjusting aperture ratio in proportion to focal length (so, equal effective aperture diameter) will increase the size of the Airy disk due to diffraction in the image formed on the sensor in that same proportion, and will also increase the size of the circle of confusion (OOF effect) at each point of the image in the same proportion. With equal pixel counts, the pixel size is also increased in the same proportion. So when you display at equal size (e.g. equal PPI for sensors of equal pixel count), everything comes out the same size!

The two imperfections I mentioned above tend to have opposite effects: lens abberations will tend to be better controlled with a larger format due to the higher f-stops used; macro or very close-up photography can work better with a smaller format and smaller sensels, because the lower magnification factor (from subject to image size on sensor) has some optical advantages.


P. S. On a personal note, the latter is why, as an enthusiast of nature close-ups, I want my pixels as small as possible, even if that means "excessively many" on a big sensor from which I crop heavily. After all, that stuff is done stopped well down, so that lens aberrations are not much of a factor. It helps that Olympus has a strong tradition of macro lenses, making its 4/3” formt digital systems a good fit for me.

[BernardLanguillier]

When the sensor can be given "full exposure", a larger sensor typically has a greater total well capacity (electons per sensel times sensels per sensor) and so can count more electons and get a better ratio of signal to shot noise, and in all but extremely deep shadows, this should win out over read noise and give a better SNR. The only trade-off for tue laeger format is either a longer exposure time (if for example you equalize DOF by using a higher f-stop in the larger format, or the larger format lenses limit you to a higher minimum f-stop), or less DOF (if for example you use equal f-stop).

And even when the tecnhological gap between modern CMOS sensors and CCDs is taken into account, I am fairly sure that CCDs can count roughly as many electons per unit area, and so win on maximum total counts of electons and thus of photons. It even more clear that the best 35mm formats sensors win over smaller formats for shot noise control at minimum EI.

OK, so we are saying that the amount of photons reaching the sensor is the same regardless of the sensor size, but that larger sensor have a greater well capacity.

I can understand that. 

Cheers,
Bernard

[Petrus]

We need a new maximum sharpness nature photography group, called F/6.3 ...

[Christoph C. Feldhaim]

How many times does this myth have to be debunked?!

The truth is that once you need to use a small aperture (high aperture ratio) in order to get enough depth of field, the effect of difraction is the same in any format, so the degree of the problem of balancing DOF againsts diffraction is almost entirely dependent on the image resolution that you are aiming for, or loosely speaking, on the pixel count.

Your mistake is comparing at equal f-stop, forgetting that with a larger sensor amd thus a larger focal length needed to get the same composition, the DOF is less in the larger format in proportion to focal length/format size. So tomget equal DOF, the large format needs to increase f-stop inroportion to focal length, which increases diffraction effects so that on same sized prints, both diffraction effects and OOF effects are equal.

Smaller formats have a disadvantage if one is seeking so much resolution that the f-stop needed to control diffrcation in a smaller format is so low rhat lens abberations become a significant factor. With 36MP in 35mm format, f/8 is fine for avoiding any problems, so we are not there yet. If you wish to avoid going below f/5.6, the resolution limit would be reached by about 120MP. Even in tiny 4/3" format, where many lenses have resolution sweet spot at f/4 or lower, the limits due to diffraction would. Ot be hit until about 60MP. In each case, I expect that other factors will limit resolution before diffraction does.

Seems this is the answer to a question that was plagueing me long since I didn't find yet time to do the math properly.

So - do I get it right like this? :
1. The F-Stop to reach a certain DoF at a certain viewing angle is linear to the sensor size as well as the diameter of the Airy disc so, that it all is basically geometrical linear and equals out.
2. The remaining problems for the smaller formats then lie in different fields like:
- DR due to smaller pixels
- More stress on the MTF of the lenses and system tolerances which need higher precision with miniaturization

Correct so or did I miss something?

[BJL]

We need a new maximum sharpness nature photography group, called F/6.3 ...

Maybe we have already had it for a long time, without knowing. The f/64 group was working with 10“x8" format, and when you scale down by a factor of 8 to 35mm film format to get equally great DOF, you get a rather familiar f/8.

But every doubling of pixel count increases linear resolution by a factor of sqrt(2) or "one stop", so equal sharpness on prints that are larger (equal PPI) and/or scrutinzed more carefully pushes the upper f-stop limit from diffraction down one stop and the lower f-stop limit to control OOF effects equally at closer scrutiny up by one stop, so do we go to f/5.6 or to f/11?! For the sharp, high DOF spirit of the f/64 movement, the shift might be towards f/11, but arguably, what matters most is still the DOF seen in viewing of the whole image, still at "normal" viewing distance, for which the same f/8 still works as well --- full sharpness under closer viewing of details is just limited to a smaller range of the details in the image, for elements of the scene closer to the plane of exact focus.


P. S. In fact, through my years of using 4/3" format, down by another factor of 2, and caring about my nature photography getting things in focus vastly more than deliberately throwing things out of focus, my battle cry has been
"f/4 and be there ... and with a camera small enough for it to be there too".

[BJL]

Seems this is the answer to a question that was plagueing me long since I didn't find yet time to do the math properly.

So - do I get it right like this? :
1. The F-Stop to reach a certain DoF at a certain viewing angle is linear to the sensor size as well as the diameter of the Airy disc so, that it all is basically geometrical linear and equals out.
2. The remaining problems for the smaller formats then lie in different fields like:
- DR due to smaller pixels
- More stress on the MTF of the lenses and system tolerances which need higher precision with miniaturization

Correct so or did I miss something?

Yes, that is my understanding, having dug through the formulas of lens optics and sensor electonics.

[FMueller]

I'm curious about other thoughts...

Nemo

My head is about to explode over this. I hire out my pixel peeping to Lloyd Chambers http://diglloyd.com/ and he is all over the deficiencies of the Canon 5d3 sensor. A logical outcome from all this pixel peeping  would be to get all my Canon gear on e-bay before the rest of you realize what junk all that stuff is.    He has some examples up on his review site pointing out the deficiencies but I must be dumb because I don't see the glaring "pattern noise" faults that he does.

But I upgraded my DSLR from a Canon 40D to a 5D3 anyway, I skipped the 5D2 for a number of reasons,among those, that I was focused on working with my M9. Like most RF users, I also keep a DSLR around for different types of "work"(this is my avocation, not my vocation, hence the quotes around work).

Oh, I did mention that I hire out my pixel peeping to Lloyd Chambers. Well, Llloyd also raked the M9 for its noise characteristics.  And then he went and spent close to $30,000 of his own money on an M9, a noctilux, a 50 and a 35 summilux and others. Then he upgraded to the M9P! A cosmetic upgrade...  All of this is a reminder that Lloyd makes a living off pixel peeping. He is being paid to split hairs.

The difference between a good image and great image will never be the hairsplitting differences between a Canon, Nikon, Pentax, Leica, etc... even though it may be interesting to banter this about when you aren't photographing, post processing, printing or presenting your work.

All this teeth gnashing has hit a high crescendo with the 5d3 and the d800/e and I think it is precisely because the technology is maturing and each new release doesn't give us the obvious improvements over the last model that we've become accustomed to since the onset of the digital age.

So, I read an interview of David Burnett http://tinyurl.com/cxr7fve I went to my bookshelf and pulled out Sam Abell's "The life of a photograph" and realized that the difference between my M9, my 5d3 or a d800 doesn't make one bit of difference.  But you knew that, I knew that, we all know that, we just forget sometimes. I'm going to shoot more and produce more, I'm going to let everyone else worry about the Canon/Nikon fan wars....lest I fall hopelessly behind.

[ejmartin]

Cramming more tiny pixels into a 35mm sized sensor makes the diffraction problem bigger. 22 MPix FF sensor already starts to loose resolution at f:8. MF sized sensor has f:8 as the diffraction limit at 60 MPix. Besides bigger pixels have less noise, better high ISO. For landscape photography sharp, small sensors are a problem as stopping down brings the effective resolution to around 5 MPix only (f:16-22 with FF sensor). With larger sensor with same MPix you can stop down more, which compensates most of the lost DOF when moving to bigger size sensor (but not all).

Comparing RAWs at f:16 means comparing two 7 MPix images, no matter what the original pixel count is on those two FF cameras.

Here is a graph of resolution of the D3 and D3x as a function of aperture with a typical lens (from a post at DPReview, taken from DxO data, IIRC):



Note a few things:
1. There is no aperture at which the sensor with fewer pixels has more, or even as good, resolution.
2. If you require a certain number of lp/mm resolution, the camera with more pixels will provide it over a larger range of apertures.
3. The resolution curve goes up as the lens is stopped down at small f/ratio due to lens aberrations; it goes down at large f/ratio due to diffraction.  However, the point at which resolution starts to decrease due to diffraction is the same independent of the number of pixels (beyond f/5.6 in this case).  Diffraction does not 'set in earlier' with more pixels.
4. There are diminishing returns at higher f/ratio, but even at f/16 the camera with 40% higher linear pixel count holds about 15% higher resolution.

So just how has the sensor with more pixels 'made the diffraction problem bigger'?

[Ray]

Here is a graph of resolution of the D3 and D3x as a function of aperture with a typical lens (from a post at DPReview, taken from DxO data, IIRC):

Graphs are wonderful things, aren't they!  They can summarise a situation so clearly.

A few years ago I went to the trouble of comparing the resolution of my 10mp Canon 40D with my upgraded 15mp Canon 50D, as a result of all the talk on internet forums about the increased pixel count of the 50D being of little use due to the effects of diffraction.

The increase in pixel numbers of the 50D, compared with the 40D, is only 50% so the resolution differences at various f stops may not be as great as the differences between the D3 and D3x, as shown on the graph. However, I find it meaningful to consider that these cropped format sensors, if full-frame, would be 26mp and 39mp (approximately) and therefore relevant to the situation comparing a D3X with a D800.

To refresh my memory I've dug out these images, which I recorded on a DVD at the time.  I'm surprised to see what appears to be moire or aliasing artifacts on the 40D shot of the banknote at F8. Now I don't happen to have a $50 bill in my wallet at present to check whether or not those broad, diagonal stripes on the background to the immediate left and right of the face, actually exist on the banknote, but I'm pretty sure they don't. I believe they're artifacts resulting from an AA filter which isn't strong enough.

When testing camera or lens qualities, I always prefer to let my eyes be the judge. I deliberately positioned the banknote at a distance which would make the text almost illegible, because those are the sorts of conditions when resolution differences can be meaningful. If one is comparing lens or sensor resolution and the differences are such that with one system a particular text is illegible, or almost illegible, but with the other system the same text is clearly legible, then there can be no doubt which system has the better resolution.

In revisiting these images, I see also that what I'd forgotten is that the 50D at F16 appears to have approximately the same resolution as the 40D at F8. In fact, in some respects the 50D image appears to be very slightly more detailed. The fact that aliasing artifacts are present in the 40D shot would indicate that focussing was 'spot on'.

For those who may be a bit concerned about my methodology, a tripod, remote release and LiveView was used with both cameras. The lens was the Canon 50/1.4.

[K.C.]

My head is about to explode over this.

There's a cure for threads like this. You just go out and shoot with what you've got and enjoy it.

[Petrus]

Here is a graph of resolution of the D3 and D3x as a function of aperture with a typical lens (from a post at DPReview, taken from DxO data, IIRC):



So just how has the sensor with more pixels 'made the diffraction problem bigger'?

It is perfectly true that the sensor with less pixels never resolves better than the sharper sensor. What happens, though, it that the sharper (more MP) sensor starts to suffer form diffraction EARLIER than the less MP sensor of the same size. You can see that from the graph also if you interpolate the graphs a bit (lesser MP sensor sharpness peaks later than the other). The only thing that means is that you are not getting the resolution advertised unless you are careful with that aperture ring. You are still getting sharper shots than the fella with less-pixel body, that remains true, or at least there is a possibility for that. I have never claimed that lesser MP sensor would take sharper shots, other things being equal.

Sorry to have caused needless posts by my unfortunate choice of words.

One thing worth noting is that to really be able to objectively compare different sized and different pixel density sensors we would need sensor which differ in only one aspect at the time. Comparing different makes of camera with different generation electronics and software bring in too many variables which are difficult to filter out if only theoretical results are looked for.