Megapixel Count, Megapixel Density x Camera Size

[joofa]

These conditions completely negate and cancel the lower shot noise advantage of the larger sensor.

Please think carefully as you seem to be ignoring the signal and the effect of pixel size.

Joofa

[Ray]

Please think carefully as you seem to be ignoring the signal and the effect of pixel size.

Joofa

I'm trying . Please re-read carefully what I wrote. I'll try to expand upon it.

At the same shutter speed, same ISO and same f stop, or more briefly, at the same 'correct' exposure whatever the f stop and shutter speed may be, the larger sensor will record less shot noise as a proportion of total signal. Agreed?

Now I can't speak for your photographic concerns, but speaking for myself, shutter speed and ISO settings for a desired DoF are two major factors that I find myself frequently juggling. Of course, if the subject is a still-life and one is using a tripod, then one can always use base ISO, and shutter speed is not an issue, but those are not the circumstances of most of my photography.

It is in these circumstances, where shutter speed and DoF are an issue, that the shot-noise advantage of the larger sensor is completely negated, if one puts both cameras on an equal footing as regards shutter speed and DoF.

For example, a sensor which is double the area of another, will require a one-stop smaller aperture to record the same depth-of-field using the appropriately longer lens for the same field-of-view.

At a one-stop smaller aperture, and at the same shutter speed, the amount of light received by the larger sensor is halved. It becomes exactly the same as the light received by the smaller sensor which is half the area, and therefore the shot noise will be the same for both sensors. Is this not correct?

Now it's true that I am ignoring the role of pixel size here because I can't see how it has much bearing on the issue of shot noise, unless one wishes to make the point that, for example, one large pixel has a greater fill factor than four smaller pixels covering the same area.

However, it seems to me, if one considers the possibly greater fill factor that one larger pixel may have, compared with 4 smaller pixels covering the same area of sensor, then perhaps one should also take into consideration quantum efficiency. Where does that lead us?

It leads us to a point I tried to make in an earlier thread in which I suggested that shot noise is proportional to the amount of light impinging upon the sensor surface.

This idea was strongly refuted by certain members of the forum who are probably more technically savvy than I am. They argued that shot noise is proportional to the number of photons actually recorded.

Even though I don't understand precisely why this may be true, I had to agree on philosophical grounds that we can't know about matters, whether noise or anything else, that haven't been recorded, or can't be recorded.

There's an interesting example which I think illustrates the dilemma. The Canon 5D (first model) has a lower base ISO than the Nikon D700. According to DXOMark, base ISO is ISO 92 for the 5D and ISO 162 for the D700. The sensors are not only the same size, but have a very similar pixel count, yet at the same F/stop, T/stop and focal length, the D700 will let in about 2/3rds of a stop less light, at base ISO for full exposure, which will require a faster shutter speed.

Does the D700 therefore record a greater quantity of shot noise (in proportion to the signal) than the 5D?

[joofa]

At the same shutter speed, same ISO and same f stop, or more briefly, at the same 'correct' exposure whatever the f stop and shutter speed may be, the larger sensor will record less shot noise as a proportion of total signal. Agreed?

The issue, which is almost always ignored on this forum and dpreview, is that when you say "record" that means sampling of the signal strength on pixels, and how those samples are reconstituted for any comparison purposes with a smaller or larger sensor all depends upon how they are combined. There is little point in theoretical treatments of "light falling on a sensor" in this case. The only practical reality are the samples we have and we have to figure out our notions of signal, noise, and SNR through them. Therefore, in this combination process, issues such as pixel size become important.

Now it's true that I am ignoring the role of pixel size here because I can't see how it has much bearing on the issue of shot noise, unless one wishes to make the point that, for example, one large pixel has a greater fill factor than four smaller pixels covering the same area.

If I have a sensor that is 4x4 times larger than a given sensor, and I want to compare its image samples with the smaller sensor then, do you think the following will be equivalent:

(1) For each 4x4 block of pixels, keep the top left pixel.
(2) For each 4x4 block of pixels keep the bottom right pixel.
(3) Take an average of 4x4 pixels.
(4) Use sinc function to get a weighted average of 4x4.
(5) Use Lanczos function to get a weighted average of 4x4.

And, which of the above is equivalent to the usual mantra that "shot noise goes along as a sqrt factor of pixels combined? In any case, at best, that mantra would be only equivalent to only one of the above, so all the rest give different notions of SNR.


Sincerely,

Joofa

[Ray]

The issue, which is almost always ignored on this forum and dpreview, is that when you say "record" that means sampling of the signal strength on pixels, and how those samples are reconstituted for any comparison purposes with a smaller or larger sensor all depends upon how they are combined. There is little point in theoretical treatments of "light falling on a sensor" in this case. The only practical reality are the samples we have and we have to figure out our notions of signal, noise, and SNR through them. Therefore, in this combination process, issues such as pixel size become important.


If I have a sensor that is 4x4 times larger than a given sensor, and I want to compare its image samples with the smaller sensor then, do you think the following will be equivalent:

(1) For each 4x4 block of pixels, keep the top left pixel.
(2) For each 4x4 block of pixels keep the bottom right pixel.
(3) Take an average of 4x4 pixels.
(4) Use sinc function to get a weighted average of 4x4.
(5) Use Lanczos function to get a weighted average of 4x4.

And, which of the above is equivalent to the usual mantra that "shot noise goes along as a sqrt factor of pixels combined? In any case, at best, that mantra would be only equivalent to only one of the above, so all the rest give different notions of SNR.


Sincerely,

Joofa

I'm not sure what you are getting at, Joofa. My understanding is that shot noise is equal to the sqrt of the total number of photons recorded, whether they are recorded by small pixels or big pixels.

It's the number of photons sampled or recorded that counts.

If one big pixel, that occupies 4x the space on the sensor as another smaller pixel, is able to record more than 4x the number of photons as one of the smaller pixels, because it has greater than 4x the fill factor, or greater than 4x the full-well capacity of one smaller pixel, then the shot noise recorded by the entire sensor consisting of big pixels, will be lower as a proportion of the total signal.

So I concede the point that a sensor consisting of larger pixels on the same size sensor as another consisting of smaller pixels, could potentially exhibit less shot noise in terms of SNR.

However, depending on the progress and direction of technological development, the 4 smaller pixels, due perhaps to greater quantum efficiency, or differences in the transmissive qualities of the CFA, or a narrowing of the spaces between each pixel, could also potentially exhibit less shot noise than an older sensor consisting of larger pixels.

Since the technological development of sensors seems to be in the direction of smaller pixels of higher performance, I get the impression that pixel size is not nearly as significant as sensor size in determining shot noise.

However, to add to the confusion I see another aspect of this. Shot noise is just one component of total noise. It's quite conceivable that a larger sensor which could potentially record a lower component of shot noise than another smaller sensor, may actually record higher quantities of other types of noise which make the lower shot noise irrelevant because it's below the noise floor.

I'm presuming that any shot noise, whether big or small, which is below a sensor's noise floor, is irrelevant.

In other words shot noise is the only noise we should ever be able see in an ideal camera, when the signal is low, as in the deep shadows, at base ISO.

Has the D7000 or K5 reached this exalted level, perhaps? 

[joofa]

Hi Ray,

The point was that the usual relation for SNR as given by using shot noise formula directly will only hold in binning a constant signal. In regular images (1) the intensity is not uniform, and, (2) hardly one would use binning in a real resampling operation. So too much attahment with SNR through sqrt is not always correct, at least theoretically.

Joofa

[Ray]

Hi Ray,

The point was that the usual relation for SNR as given by using shot noise formula directly will only hold in binning a constant signal. In regular images (1) the intensity is not uniform, and, (2) hardly one would use binning in a real resampling operation. So too much attahment with SNR through sqrt is not always correct, at least theoretically.

Joofa

Understood! But the randomness of the signal which results in what we call shot noise has nothing whatsoever to do with the type of camera used, the size of the sensor or its pixels.

Whilst the degree of shot noise recorded is affected by the type of camera used, the exposure and the size of the sensor etc, the nature of the randomness which causes the shot noise is independent of the camera.

If I set up any camera on a tripod and take multiple shots of any still life which is uniformly lit, or as uniformly lit as it's possible to make it, I will not get equal exposures of the identical pixels in each shot. In fact it's quite surprising to observe the dramatic reduction in noise when a number of such images are stacked in Photoshop Extended resulting in a composite image consisting of the best exposed pixels from each frame.

However, the average of the total shot noise in each frame will vary much less than the exposure variation of individual pixels of identical location in each frame.

When comparing equal size sensors which have unequal size pixels, say a 6mp 35mm sensor with a 24mp 35mm sensor, both with equal base ISO sensitivity, the question we should ask is, "Is there any reason to suppose that an exposure with the 6mp camera, at a precise moment in time, unique in the history of the universe, will record less shot noise than the 24mp camera positioned at exactly the same location?"

Theoretically, if the ISO and the exposure duration is the same for both cameras, each group of 4 pixels on the 24mp sensor, occupying the same space and location as the equivalent single, large pixel on the 6mp sensor, will record the same number of photons, at precisely the same point in time and same duration of time. Is this not correct?

Of course, it's understood we can't have two cameras occupying the same space at the same time, so this is theoretical. 

[joofa]

But the randomness of the signal which results in what we call shot noise has nothing whatsoever to do with the type of camera used, the size of the sensor or its pixels.

How that shot noise is packaged for comparison purposes, when you want to compare the larger and smaller sensor image data, depends upon the number of samples, so that is why I tend to think that pixel size is important. Please see my example above the with the 5 options of going from a 4x4 times larger sensor to the smaller sensor.

Joofa

[Ray]

How that shot noise is packaged for comparison purposes, when you want to compare the larger and smaller sensor image data, depends upon the number of samples, so that is why I tend to think that pixel size is important. Please see my example above the with the 5 options of going from a 4x4 times larger sensor to the smaller sensor.

Joofa

Joofa,
If I understand you correctly, you're saying that the downsampling algorithms leave something to be desired and that 4 pixels binned at the time of exposure will result in less noise than 4 pixels unbinned and later downsampled in post-processing.

I can see that this may be true, but I'm having difficulty in relating it to shot noise. As I understand, if all else is equal (which it rarely is, of course), the read noise for one large pixel, consisting of 4 binned pixels, will be less than 4x the read noise of one small pixel. Other noise up the processing chain may also be less, proportional to the signal.  Therfore, from the perspective of noise in general, but noise which has been created by the camera, larger pixels have an advantage.

But shot noise is not created by the camera, so is in a different category.

Shot noise as I understand it, is due to an unavoidable randomness of the arrival of photons, which is simply part of the fabric of reality.

Getting back to my imaginary scenario of two cameras occupying the same position at the same time, a 6mp and a 24mp FF 35mm camera. If we imagine that exactly 1,000 photons have been sampled by a specific large pixel from the 6mp camera, and that exactly 1,000 photons would have been sampled by the block of 4 smaller pixels occupying exactly the same space at the same time, then the shot noise for the 24mp camera, at that location of 4 pixels, will be defined by the degree of variation of signal strength amongst those 4 pixels.

The number of photons actually sampled will be different for each pixel in the block of 4, even though the light source is apparently uniform. In total, the number of photons sampled will be 1,000, but the top left pixel might have sampled 230 photons, the bottom right, 240 photons, and the top right 270 photons etc.

The main point here is that it's the variation that is the cause of the noise. Eliminate the variation and you've eliminated the noise.

Now, I'm not an expert on the various downsampling methods that are available. I'm not terribly concerned about minute differences between different RAW converters, different sharpening algorithms, or different resampling algorithms, but looking at your 5 descriptions of downsampling methods, it would seem to me that any average of the 4 pixels would completely equalise shot noise in relation to the larger pixel.

(1) For each 4x4 block of pixels, keep the top left pixel.
(2) For each 4x4 block of pixels keep the bottom right pixel.
(3) Take an average of 4x4 pixels.
(4) Use sinc function to get a weighted average of 4x4.
(5) Use Lanczos function to get a weighted average of 4x4.

[joofa]

Ray,

The point is that for the treatment we are discussing it does not matter that shot noise is in the measure of number of photons collected. The thing that matters is that the recorded value of a pixel has that noise incorporated. But, any recombination of these noisy values coming from neighboring smaller pixels to compare to the single valued obtained from a larger pixel occupying the same spatial extent does not necessarily give the same relation of signal to noise unless the recombination method is binng.

Joofa

[Peter_DL]

... the recorded value of a pixel has that noise incorporated. But, any recombination of these noisy values coming from neighboring smaller pixels to compare to the single valued obtained from a larger pixel occupying the same spatial extent does not necessarily give the same relation of signal to noise unless the recombination method is binng.

So the Megapixel density [MP/cm2] counts (?).
It was an initial assumpion, and the rough correlation found with the DxO overall Sensor Mark seemed to confirm it.

Peter

--

[joofa]

So the Megapixel density [MP/cm2] counts (?).
It was an initial assumpion, and the rough correlation found with the DxO overall Sensor Mark seemed to confirm it.

Peter

--

Hi,

This is what I tend to think so. The proponents of "sensor size is all that matters" may not agree with this thing. But, as I have pointed out before, we are not doing a theoretical analysis of "light falling on a sensor". In practical terms what we have are samples and any comparison of samples from a different sized sensor has to be cognizant of the method of comparison.

But I think DxO does not figure out the relationship of SNR in resampling to account for sensor size difference properly. What they, and many people, ignore is that a signal is also blurred in a resampling operation. But DxO uses a simple formula which is agnostic to signal blurring.

Sincerely,

Joofa

[Ray]

This is what I tend to think so. The proponents of "sensor size is all that matters" may not agree with this thing. But, as I have pointed out before, we are not doing a theoretical analysis of "light falling on a sensor". In practical terms what we have are samples and any comparison of samples from a different sized sensor has to be cognizant of the method of comparison.

Joofa,
I don't necessarily agree with the principle that sensor size is all that matters. The larger pixel may have a fill-factor advantage, and I believe the CCD design may also allow for a greater fill factor compared with the CMOS design which may include additional wires and transistors in areas that otherwise could collect photons.

If everything were the same, apart from pixel size, a block of 4 small pixels occupying the same space as one large pixel would probably not be able to sample the same number of photons as the single larger pixel, during a given exposure, because there are divisions and borders between the pixels comprising the block of four, where photons would be falling on infertile ground.

I'm presuming that a block of 4 smaller pixels would be able to sample the same number of photons as the single larger pixel, only if the smaller pixel had some technological advantage such as greater quantum efficiency.

But I think DxO does not figure out the relationship of SNR in resampling to account for sensor size difference properly. What they, and many people, ignore is that a signal is also blurred in a resampling operation. But DxO uses a simple formula which is agnostic to signal blurring.

Are we talking about upsampling or downsampling? Are we talking about minute variations of the order of rounding errors during mathematically intense calculations?

How does your argument relate to Michael's comparison of a P45+ image with a G10 shot of the same scene, both at A3+ print size, which were indistinguishable, apart from DoF differences.

[Peter_DL]

The proponents of "sensor size is all that matters" may not agree with this thing.

In the parallel discussion on the DxO sensor Mark it was stated by Peter van den Hamer:
"You still need at least 3 yardsticks - just for image quality alone: noise data (e.g. DxO Mark Sensor), resolution (sensor MPixels), and the impact of lens quality ".

Now when we pull out the absolute MegaPixel count from the sensor size (as well),
Sensor Size = Megapixel [MP] x 1/ Megapixel Density [MP/cm2].

Hence, a rough correlation can be found between the DxO Sensor Mark and the inverse Megapixel Density (see attachment).
A 'log' was included just for the purpose of illustration i.e. getting it (visually) a bit more linear.

How does your argument relate to Michael's comparison of a P45+ image with a G10 shot of the same scene, both at A3+ print size, which were indistinguishable, apart from DoF differences.

The DxO Overall Sensor Mark is dependent on the High ISO performance.
Michael's finding probably refers to Low ISO where P&S cameras shine and differences to larger sensors are less pronounced.

Best regards, Peter

--

[Ray]

The DxO Overall Sensor Mark is dependent on the High ISO performance.

Peter,
I don't pay much attention to the over all DXO score. A camera is a tool, and I want to know how it performs under specific circumstances compared with other cameras used in the same circumstances.

The more detailed DXOMark graphs show how specific attributes of DR, SNR, Tonal Range etc vary amongst different camera models, both at normalised print sizes and at the pixel level, and at various ISO settings

That's enough information for me, together with other basic knowledge about focal length and DoF equivalents (for same FoV) with different size sensors, to get a fairly accurate idea of how a camera will perform relative to another camera, under specific circumstances where certain combinations of F/stop for DoF, shutter speed for freezing of subject movement, and consequent increase in ISO (or underexposure) will apply.

Michael's finding probably refers to Low ISO where P&S cameras shine and differences to larger sensors are less pronounced.

Not at all! Both DBs and P&S cameras have the similar characteristic of being essentially one-ISO cameras. Their best performance tends to be at base ISO whatever the underexposure. That is, there's little IQ advantage in increasing ISO to use faster shutter speeds, and perhaps none whatsoever, apart from a brighter review image on the camera's LCD screen.

Canon DSLRs have always had an advantage in this respect, but currently overtaken by Nikon. However, the Nikon D7000 is now in the same category as many P&S and MFDB cameras in this respect, ie. no IQ advantage in increasing ISO to accommodate a faster shutter speed.

But of course, Nikon have not achieved this lofty status by downgrading performance at higher ISOs, but by upgrading performance at lower ISOs.

What I find fascinating about Michael's comparison of the G10 and the P45+, is not just the surprising result when images from these two cameras were normalised at A3+ size, but how potentially disastrous the result would have been for the P45+ if both DoF and shutter speed had been equalised in both shots.

In defense of Michael, this comparison was an after-thought, on the spur of the moment. The P45+ was on a solid tripod. Exposure was 1 sec at F11 and ISO 50.

The G10 at F3.5 and ISO 80 required a considerably faster shutter speed of 1/8th which, with the help of image stabilisation, didn't need a tripod, although the G10 was rested on top of the tripod base to help reduce camera shake.

Here's the article: http://luminous-landscape.com/reviews/kidding.shtml

Now, one can't help wondering what such a comparison would have been like, had Michael equalised both DoF and shutter speed for these two shots.

I can understand why he didn't, but supposing he had? What would the result have been, apart from lots of angry emails from MFDB owners?

Let's look at the bare facts which can be understood by anyone with even the most basic knowledge of mathematics.

The P45+ sensor is 36mm x 48mm. The G10 sensor is 5.7mm x 7.6mm. Both sensors are the same aspect ratio so that makes it less contentious when adjusting FL and F/stop for equal FoV and equal DoF.

The multiplier for equivalent F/stop and FL is either 36/5.7 or 48/7.6, which in both cases equals 6.3x.

When the G10 is used at F3.5, then for equivalent DoF the P45+ should be used at F3.5x6.3=F22.

Now let's consider what the settings would have been if Michael had equalised DoF in both shots.

Instead of F11 and 1 sec exposure at ISO 50, it would have been F22 and 4 secs exposure at ISO 50 with the P45+.

Supposing Michael had also equalized shutter speed, to get 1/8th sec exposure with the P45+, because the subject is not totally still?

We have to increase ISO by at least 5 stops, or underexpose by at least 5 stops, and that's not taking into consideration the difference between ISO 50 and ISO 80, so for a full ETTR exposure for both cameras, the difference is actually more than 5 stops.

But, let's give the advantage to the P45+ and ignore that discrepancy.

5 stops up from ISO 50 is ISO 1600, or 5 stops underexposure at base ISO, or one stop underexposure at the maximum ISO setting of 800 for the P45+. Take your pick.

Now, let's look at the comparative graphs for the P45+ and the G10 for the major parameters measured, SNR, DR, Tonal Range and Color Sensitivity, at these equivalent ISO settings of ISO 80 for the G10 and ISO 1600-3200 (extrapolated) for the P45+.

Well!! One might well be flabbergasted! It's no wonder Michael toned down his comparison. On this completely equal playing field, the P45+ fails miserably. It doesn't even qualify for the competition.

Check out the results for yourself. You don't need me to spoon-feed you. (But maybe you do  )

Of course, it has to be said, for the benefit of anyone who is confused, that on a different playing field, in different circumstances, the G10 would not even qualify. The P45+ would reign supreme, by a large margin.

[ErikKaffehr]

Ray,

I don't argue with your reasoning. You certainly make several very good points. There has always been an argument for smaller formats, essentially based on DoF, diffraction and the small format being good enough.

The revolution we have now is that smaller formats also turn out very good. Everything has been pushed a bit. Michael compares the new 80 MPixel backs to 8x10" and this may be less far fetched than it may seem. The increased quality also pushes limitations. I'm pretty sure that 24 MP is hard to reach on the D3X without tripod, mirror lockup and dead on focus. I'd suggest that the same applies to 16 MP on a Nikon D7000.

I'd suggest that MFDBs are mostly used in benign conditions, with camera on tripod or using flash. A tripod often helps, the photographer doesn't need to focus on aiming the camera. It's certainly possible to use an MFDB for handheld shooting but all the pixels may not be fully utilized.

You are certainly right about DoF, but I'd suggest that when we use larger formats we essentially need to live with very little DoF. An owner of a Pentax 645D complained that he had infinity in focus but not a boat in the foreground that was say 200 m from the camera. This was taken with a 150 mm lens at f/9.5, that would take care of DoF for sure?

In my article here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/49-dof-in-digital-pictures it was demonstrated that even small amount of defocus are clearly visible. This test was done with a 24.6 MP DSLR having the same sensor pitch as the aforementioned P645D using 150 mm focal length. On page two of the article the effects of defocus are studied with a 16 MP APS-C sensor and an 100/2.8 macro lens still at 3.0 meter. I plan to add more data to that article later on.

That article also shows the effects of diffraction.

What can be seen from this testing is that when we go to small pixels and demand optimal sharpness the DoF is very thin, it could be argued it's almost non existing. My conclusion is essentially that we either:

- Live with a small zone of absolute focus
- Stop down a lot, thus compromising resolution
- Achieve extended DoF using Scheimpflug or extending DoF by combining several images (as dicussed here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/29-handling-the-dof-trap )

Best regards
Erik

Peter,
I don't pay much attention to the over all DXO score. A camera is a tool, and I want to know how it performs under specific circumstances compared with other cameras used in the same circumstances.

The more detailed DXOMark graphs show how specific attributes of DR, SNR, Tonal Range etc vary amongst different camera models, both at normalised print sizes and at the pixel level, and at various ISO settings

That's enough information for me, together with other basic knowledge about focal length and DoF equivalents (for same FoV) with different size sensors, to get a fairly accurate idea of how a camera will perform relative to another camera, under specific circumstances where certain combinations of F/stop for DoF, shutter speed for freezing of subject movement, and consequent increase in ISO (or underexposure) will apply.

Not at all! Both DBs and P&S cameras have the similar characteristic of being essentially one-ISO cameras. Their best performance tends to be at base ISO whatever the underexposure. That is, there's little IQ advantage in increasing ISO to use faster shutter speeds, and perhaps none whatsoever, apart from a brighter review image on the camera's LCD screen.

Canon DSLRs have always had an advantage in this respect, but currently overtaken by Nikon. However, the Nikon D7000 is now in the same category as many P&S and MFDB cameras in this respect, ie. no IQ advantage in increasing ISO to accommodate a faster shutter speed.

But of course, Nikon have not achieved this lofty status by downgrading performance at higher ISOs, but by upgrading performance at lower ISOs.

What I find fascinating about Michael's comparison of the G10 and the P45+, is not just the surprising result when images from these two cameras were normalised at A3+ size, but how potentially disastrous the result would have been for the P45+ if both DoF and shutter speed had been equalised in both shots.

In defense of Michael, this comparison was an after-thought, on the spur of the moment. The P45+ was on a solid tripod. Exposure was 1 sec at F11 and ISO 50.

The G10 at F3.5 and ISO 80 required a considerably faster shutter speed of 1/8th which, with the help of image stabilisation, didn't need a tripod, although the G10 was rested on top of the tripod base to help reduce camera shake.

Here's the article: http://luminous-landscape.com/reviews/kidding.shtml

Now, one can't help wondering what such a comparison would have been like, had Michael equalised both DoF and shutter speed for these two shots.

I can understand why he didn't, but supposing he had? What would the result have been, apart from lots of angry emails from MFDB owners?

Let's look at the bare facts which can be understood by anyone with even the most basic knowledge of mathematics.

The P45+ sensor is 36mm x 48mm. The G10 sensor is 5.7mm x 7.6mm. Both sensors are the same aspect ratio so that makes it less contentious when adjusting FL and F/stop for equal FoV and equal DoF.

The multiplier for equivalent F/stop and FL is either 36/5.7 or 48/7.6, which in both cases equals 6.3x.

When the G10 is used at F3.5, then for equivalent DoF the P45+ should be used at F3.5x6.3=F22.

Now let's consider what the settings would have been if Michael had equalised DoF in both shots.

Instead of F11 and 1 sec exposure at ISO 50, it would have been F22 and 4 secs exposure at ISO 50 with the P45+.

Supposing Michael had also equalized shutter speed, to get 1/8th sec exposure with the P45+, because the subject is not totally still?

We have to increase ISO by at least 5 stops, or underexpose by at least 5 stops, and that's not taking into consideration the difference between ISO 50 and ISO 80, so for a full ETTR exposure for both cameras, the difference is actually more than 5 stops.

But, let's give the advantage to the P45+ and ignore that discrepancy.

5 stops up from ISO 50 is ISO 1600, or 5 stops underexposure at base ISO, or one stop underexposure at the maximum ISO setting of 800 for the P45+. Take your pick.

Now, let's look at the comparative graphs for the P45+ and the G10 for the major parameters measured, SNR, DR, Tonal Range and Color Sensitivity, at these equivalent ISO settings of ISO 80 for the G10 and ISO 1600-3200 (extrapolated) for the P45+.

Well!! One might well be flabbergasted! It's no wonder Michael toned down his comparison. On this completely equal playing field, the P45+ fails miserably. It doesn't even qualify for the competition.

Check out the results for yourself. You don't need me to spoon-feed you. (But maybe you do  )

Of course, it has to be said, for the benefit of anyone who is confused, that on a different playing field, in different circumstances, the G10 would not even qualify. The P45+ would reign supreme, by a large margin.

[Peter_DL]

Ray,

Ok, get your point.
Many thanks for explanation.
Lens stabilization contributes a lot to stay at expedient base ISO with the G10, as far as I can tell. But then, even at base ISO the SNR 18% noise level with a G10 is just 'orange' and worse compared to larger sensor cameras – although with the latter it may not be feasible without a tripod, particularly when a same DoF is to be met.
 
Still I'd claim that the G10 camera size is a bit too large for the given sensor size (other P&S cameras of same or similar sensor quality like the S95 or LX5 are more pocketable). Or, the camera size is ok but the sensor could be somewhat larger, maybe 12 to 14 MP at an lower MP density. A next 'G14' may face up the challenge and move in the direction of m4/3. Let's see.

Peter

What I find fascinating about Michael's comparison of the G10 and the P45+, is not just the surprising result when images from these two cameras were normalised at A3+ size, but how potentially disastrous the result would have been for the P45+ if both DoF and shutter speed had been equalised in both shots.

In defense of Michael, this comparison was an after-thought, on the spur of the moment. The P45+ was on a solid tripod. Exposure was 1 sec at F11 and ISO 50.

The G10 at F3.5 and ISO 80 required a considerably faster shutter speed of 1/8th which, with the help of image stabilisation, didn't need a tripod, although the G10 was rested on top of the tripod base to help reduce camera shake.

Here's the article: http://luminous-landscape.com/reviews/kidding.shtml

Now, one can't help wondering what such a comparison would have been like, had Michael equalised both DoF and shutter speed for these two shots.

I can understand why he didn't, but supposing he had? What would the result have been, apart from lots of angry emails from MFDB owners?

Let's look at the bare facts which can be understood by anyone with even the most basic knowledge of mathematics.

The P45+ sensor is 36mm x 48mm. The G10 sensor is 5.7mm x 7.6mm. Both sensors are the same aspect ratio so that makes it less contentious when adjusting FL and F/stop for equal FoV and equal DoF.

The multiplier for equivalent F/stop and FL is either 36/5.7 or 48/7.6, which in both cases equals 6.3x.

When the G10 is used at F3.5, then for equivalent DoF the P45+ should be used at F3.5x6.3=F22.

Now let's consider what the settings would have been if Michael had equalised DoF in both shots.

Instead of F11 and 1 sec exposure at ISO 50, it would have been F22 and 4 secs exposure at ISO 50 with the P45+.

Supposing Michael had also equalized shutter speed, to get 1/8th sec exposure with the P45+, because the subject is not totally still?

We have to increase ISO by at least 5 stops, or underexpose by at least 5 stops, and that's not taking into consideration the difference between ISO 50 and ISO 80, so for a full ETTR exposure for both cameras, the difference is actually more than 5 stops.

But, let's give the advantage to the P45+ and ignore that discrepancy.

5 stops up from ISO 50 is ISO 1600, or 5 stops underexposure at base ISO, or one stop underexposure at the maximum ISO setting of 800 for the P45+. Take your pick.

Now, let's look at the comparative graphs for the P45+ and the G10 for the major parameters measured, SNR, DR, Tonal Range and Color Sensitivity, at these equivalent ISO settings of ISO 80 for the G10 and ISO 1600-3200 (extrapolated) for the P45+.

Well!! One might well be flabbergasted! It's no wonder Michael toned down his comparison. On this completely equal playing field, the P45+ fails miserably. It doesn't even qualify for the competition.

[Peter_DL]

The revolution we have now is that smaller formats also turn out very good.

One of my best investments about 10 years ago was a Leica minilux zoom (35-70 mm lens). Equipped with Full Format 35 mm film, resulting slides were often enough undistinguishable from a SLR + prime lens. Not sure why it is taking 'digital' so long to come to a similar break even. Is it just me seeing it so (?).

Best regards, Peter

--

[Plekto]

The large problem with digital is the sensor pattern that they use.  It's basically the same that you see in use for magazine printing and the like - a grid of red, blue, and green dots.  Many cameras have horrible software and do a fairly bad job of making a clean image out of all of that.  The best 35mm film with great lenses has about 12MP worth of information comparable to a digital camera.   But, often, it still has slightly better color and dynamic range unless you're looking at a professional camera.  ie - a 12MP pocket camera will generally suffer quite a bit, just like how sticking Velvia in a 40 year old rangefinder is likely to produce less than perfect results.

The general rule of thumb is to just buy a 16-24MP DSLR, so that even with the basic lenses, you'll still outclass film by a large margin. 

[Ray]

I'm pretty sure that 24 MP is hard to reach on the D3X without tripod, mirror lockup and dead on focus. I'd suggest that the same applies to 16 MP on a Nikon D7000.

Erik,
I'm a great fan of Image Stabilisation, whether of the sensor or of the lens, and good image quality at high ISO, or even at moderately high ISO such as ISO 200 & 400.

These two factors used together should allow one to get sharp results in most situations without use of a tripod.

Most MFDBs are at a clear disadvantage in this respect, lacking both IS and good performance at high ISO.

I remember when buying my first Nikkor lens, the 14-24/2.8. I was a bit dissapointed it didn't boast VR. However, I didn't consider it too serious an issue because wide-angle lenses don't require as fast a shutter speed as longer focal lengths. Even if one's standards for a sharp, hand-held shot require shutter speeds of 1/2FL, that's still only 1/30th sec at 14mm and 1/50th at 24mm.

Initially I used my Canon 5D with this lens, then decided the lens was so good it deserved a matching body for full functionality, so I bought the Nikon D700. This proved to be an upgrade in more ways than one. The D3, almost identical to the D700, caused quite a stir when it was first released because of its superb high-ISO performance which, it was claimed, was streets ahead of any other camera.

In fact, it wasn't the high ISO which was streets ahead, but the low to moderate ISOs which were streets ahead , ie, ISO 200-800.

In other words, the D700 at a real ISO 200 not only has at least the same performance as the 5D at a real ISO 100, with respect to SNR at 18% grey, tonal range and color sensitivity, but also has a full stop better DR.

However, it gets better. Suppose I can use a 1/30th at 14mm and ISO 200, with the D700 attached to this lens. With the 5D attached to the lens, I need to boost ISO to 200 to use the same shutter speed. How does a 'real' ISO 200 on the 5D compare with a 'real' ISO 200 on the D700?

DXOMark provide the answer. About 1 stop better SNR at 18% grey; at least 1 bit better color sensitivity (which is noticeable); marginally better tonal range, and still a full stop better DR.

Now the D7000 is in a different category with its higher pixel count and smaller sensor. A 1/30th at 14mm is no longer adequate for a sharp, hand-held shot.

Let's consider how this changes the 1/FL rule which I have upgraded to 1/2FL, because my standards are high, (or because my hands are shaking - take your pick  ).

The longer the focal length, the narrower the angle of view. These two factor tend to be in  proportion with normal, rectilinear lenses, just as the shutter speed for a sharp, hand-held shot is also in proportion to the focal length, and inversely proportional to the angle of view, or FoV.

The same proportionality applies to increased pixel density. The maths is really very simple for the approximations that are relevant for practical photography. There's nothing esoteric here.

Consider a specific size of sensor with a specific number of pixels. Let's presume that one is healthy, one doesn't drink, one practices Pranayama Breathing exercises, and one is as steady as a rock.

The 1/FL rule will apply for 8"x12"prints using FF 35mm format. It might even be overkill for some of us.

For ease of calculation, let's presume that a 6mp sensor can produce an 8x12"print at 256 ppi without interpolation, and that a 24mp sensor can produce a 16"x24"print in the same circumstances, at the same ppi.

The sensor's pixel count is exactly 4x greater, and the area of the print is 4x greater. Does that mean we should use a 4x faster shutter speed to achieve the same degree of sharpness, at the pixel level?

No. If we crop the 24mp image to 6mp, we get the same FoV as we would have got with the 6mp camera used with a lens double the focal length.

The 1/FL rule that was adequate for the 6mp FF DSLR, now becomes 1/2FL for the 24mp DSLR. No big deal. No need to get one's nickers in a twist. Just be aware of it.