Canon 50D @ 15MP

[Panopeeper]

"Despite the fact that the pixels are smaller and one camera has a 10.1MP sensor and the other has a 15.1MP sensor, you'd think you'd have more noise but you have less," he said. "The end result is better image quality all around which allows us to extend the ISO range to 12800.

The 50D is able to achieve this because of new "gapless microlenses" on the image sensor which capture more light and generate less noise

This is all right, but it does not translate in higher dynamic range. Capturing more light in the same time causes a shift on the ISO scale: the number of photons required can be collected in shorter time - BUT the wells will become full sooner as well. In other words, not only the lower but the higher end of the dynamic range moves as well.

In order to increase the dynamic range, either the well capacity has to be increased or the required number of photons decreased; none of these has to do with microlenses. It is like with the aperture: a larger aperture allows capturing the required number of photons in shorter time, while at the same time saturation too will occur in shorter time.

[Tony Beach]

In order to increase the dynamic range, either the well capacity has to be increased or the required number of photons decreased; none of these has to do with microlenses.[a href=\"index.php?act=findpost&pid=217428\"][{POST_SNAPBACK}][/a]

Since there is more absolute DR than there is useful DR (in all the DSLRs I'm aware of), noise is a more limiting factor than well capacity.

[BernardLanguillier]

While the 50D looks like a VERY nice camera, I can assure you that you'll be treading into diffraction limited waters once near or past F/8 and purely from a resolution standpoint, if you shot the 40D and 50D at F/11, you'd quite likely end up with about the same overall resolution simply because of the pixel pitch of the higher MP sensor in the same APS-C space. Has nothing to do with the lens at this point.
[a href=\"index.php?act=findpost&pid=217370\"][{POST_SNAPBACK}][/a]

All true, but we will be close to that soon with FX too.

The fact is that we are at levels of resolution on "small sensors" that is such that it now clearly impossible to get both a lot of DoF and very good sharpness.

The only 2 options are:

1. T/S lenses
2. Digital DoF stacking.

I have been using 2. for years and it works a treat for landscape work.

As far as the 50D goes, I personnally welcome Canon's willingness to keep pushing the DX enveloppe further. The need to use f8 to shoot at optimal sharpness is not a problem in itself. You will still always have a better sharpness/DoF compromise with DX than with FX.

As mentioned above though, my personnal feeling is that it would have been better to push the DX enveloppe towards more DR at same pixel count, but this opinion is not derived from the fact that more pixels mean more diffraction.

Cheers,
Bernard

[Panopeeper]

Since there is more absolute DR than there is useful DR (in all the DSLRs I'm aware of), noise is a more limiting factor than well capacity.

The dynamic range, as its name suggests, has to "ends". Neither end is more limiting than the oder, for the range is the distance between these ends.

Anyway, if Canon in fact could enhance the noise characteristics of the sensor, that increases the dynamic range. However, that has nothing to do with the microlenses.

[BernardLanguillier]

Since there is more absolute DR than there is useful DR (in all the DSLRs I'm aware of), noise is a more limiting factor than well capacity.
[a href=\"index.php?act=findpost&pid=217432\"][{POST_SNAPBACK}][/a]

There should be no such things as "DR" and "useful DR". If DR is not useful then the definition of DR in terms of the level of noise in the shadows from which there is no detail is wrong.

DR is often confused with highlight rollout.

The recent sensors, at least from Nikon that I know better, are doing a much better job at having a smooth transition to blown out areas. This is not the result of more DR, but is the result of a better sensor design coupled with more bits to handle the very bright parts of the image.

DR shows mostly in the shadows of most DSLRs (the onlyt exception being the Fuji S5) in terms of how much you can lighten up the shadows without getting un-acceptable noise.

Cheers,
Bernard

[Panopeeper]

The recent sensors, at least from Nikon that I know better, are doing a much better job at having a smooth transition to blown out areas. This is not the result of more DR, but is the result of a better sensor design coupled with more bits to handle the very bright parts of the image

There is no "smooth transition" to pixel saturation. The sensors of the Nikon cameras do not behave differently from any other: the recorded pixel values are linear (if one adjusts for a floating start), up to the saturation.

However, the camera's software or the raw conversion can cause such an effect; see HTP with Canons, and I think Active Daylight is doing something similar (though it does not seem to reduce the ISO as HTP does).

DR shows mostly in the shadows of most DSLRs (the onlyt exception being the Fuji S5) in terms of how much you can lighten up the shadows without getting un-acceptable noise

When I am stating, that I am not interested for the highlights when measuring the DR, only for the very deep shadows, some people think I am joking. I need only a very dark shot (with suitable subjects) to measure the DR.

[BernardLanguillier]

There is no "smooth transition" to pixel saturation. The sensors of the Nikon cameras do not behave differently from any other: the recorded pixel values are linear (if one adjusts for a floating start), up to the saturation.
[a href=\"index.php?act=findpost&pid=217447\"][{POST_SNAPBACK}][/a]

I'll have to disagree with you on this one.

The sensor technology does clearly influence the way subtely different very bright illuminations level are translated into RGB values.

Cheers,
Bernard

[BruceHouston]

This is all right, [span style=\'font-size:14pt;line-height:100%\']but it does not translate in higher dynamic range.[/span] Capturing more light in the same time causes a shift on the ISO scale: the number of photons required can be collected in shorter time - BUT the wells will become full sooner as well. In other words, not only the lower but the higher end of the dynamic range moves as well.

In order to increase the dynamic range, either the well capacity has to be increased or the required number of photons decreased; none of these has to do with microlenses. It is like with the aperture: a larger aperture allows capturing the required number of photons in shorter time, while at the same time saturation too will occur in shorter time.
[a href=\"index.php?act=findpost&pid=217428\"][{POST_SNAPBACK}][/a]

[span style=\'font-size:11pt;line-height:100%\']Let's take it one more time, from the top, with the important portions of the Canon statement highlighed for emphasis:[/span]

Westfall said one of the keys about the 50D is its [span style=\'font-size:14pt;line-height:100%\']enhanced noise reduction [/span]which is "a stop to a stop and half better than the 40D."

"Despite the fact that the pixels are smaller and one camera has a 10.1MP sensor and the other has a 15.1MP sensor, you'd think you'd have more noise but you have less," he said. "The end result is better image quality all around which allows us to extend the ISO range to 12800.

The 50D is able to achieve this because of new [span style=\'font-size:14pt;line-height:100%\']"gapless microlenses" on the image sensor which capture more light and generate less noise[/i].[/span]

"It's the further evolution of a concept we've already explored. It's gotten to the point where the microlenses are totally gapless and larger which produces the best efficency on a per pixel basis. [span style=\'font-size:14pt;line-height:100%\']Consequently [span style=\'font-size:21pt;line-height:100%\']the light sensitive portion of the pixel has increased [/span]while the non-light sensitive portion of the pixel has been shrunk down,"[/b][/span] he said.


[span style=\'font-size:11pt;line-height:100%\']Gapless microlense => larger pixel possible => bigger well + higher SNR;

Bigger well + higher SNR => greater dynamic range;

=> Canon's new "gapless microlense" sensor (presumably) has greater dynamic range.

(Let's play "connect the dots.")[/span]

[Tony Beach]

There should be no such things as "DR" and "useful DR". If DR is not useful then the definition of DR in terms of the level of noise in the shadows from which there is no detail is wrong.[a href=\"index.php?act=findpost&pid=217445\"][{POST_SNAPBACK}][/a]

Well, different photographers define the DR they are willing to use coming from the same camera differently.  Ironically, they also define useful ISO differently -- that's because ISO and DR are essentially the same issue, which is how much noise is acceptable to you and that cannot be quantified whereas absolute DR can be quantified.

[Panopeeper]

The sensor technology does clearly influence the way subtely different very bright illuminations level are translated into RGB values

There is no such thing. The sensors' output is strictly linear (disturbed somewhat by the noise).

See the attached captures from a D3 shot of a Stouffer wedge. The strips are roughly 1/3 stop apart, this is visible on the histogram as well. The next captures show the average pixel values on a selected spot of adjacent strips. You only need to divide the pixel values to see the scheme.

Sidenote: the raw pixel values are not RGB. They could be characterized as Rgb, Grb, Brg.

I don't know why the captures do not appear in the sequence I uploaded them. The sequence of the numbered strips are 4, 5, 6, 7, starting from the brightest one.

[Panopeeper]

Let's take it one more time, from the top, with the important portions of the Canon statement highlighed for emphasis

I'm afraid you are reading too much into a propaganda communiqué. Canon and Mr. Westfall are extremely selective with any technical information. (Example: they were not prepared to reveal the way HTP is working - like it would be some information worthy for keeping secret.)

I suggest to use some reasoning. For example: by mounting a large ancilliary lens on the camera's lens' front, one could increase the amount of captured light. This would lead to an increase of the nominal ISO. Would this increase the DR as well?

Of course, the larger/more effective microlenses increase the amount of captured light. This leads to less noise compared to a shot with teh same exposure but less effective microlenses. However, again, this leads to earlier saturation as well.

[Ray]

I did not mention MP at all. In fact, higher MP is making diffraction worse. Judged from the 40D, diffraction will be strongly noticable on the 50D already at f/11.
[a href=\"index.php?act=findpost&pid=217350\"][{POST_SNAPBACK}][/a]

Gabor,
If I may say so without causing offense, I get the impression that all this pixel-peeping has resulted sometimes in your not being able to see the wood for the trees.

System resolution is generally a combination of lens resolution and film or sensor resolution. If the resolution of either one is increased, whilst keeping the other the same, the result will likely (but not necessarily) still be a modest increase in system resolution, depending on other factors such as the strength of the AA filter.

An example of this principle in its most basic form would be as follows.

If a lens at a particular aperture delivers a resolution of 50 lp/mm at 50% MTF, and the sensor also can record 50 lp/mm with a loss of 50% contrast, then the system resolution would be 1/(1/50 +1/50) = 25lp/mm at 50% MTF.

However, if we increase sensor resolution by increasing pixel count, so that the sensor is capable of recording 100 lp/mm at 50% MTF, then the system resolution using the same lens at the same aperture is given by 1/(1/100+1/50) = 33 lp/mm.

However, I believe such formulas are very rough approximations because of the role of the AA filters and demosaicing algorithms.

Nevertheless, my own tests have indicated that a 20D with with the 100-400 IS zoom at 400mm and F22 can produce marginally more detailed results than the same lens at the same aperture on the 5D (shooting the same scene from the same position). I would therefore expect that a 50D with the same lens at the same aperture of F22 would also produce marginally more detailed results than the 20D to a degree which might be significant on large prints.

However, the difference between the 50D and 450D in this respect will likely be inconsequential. BJL mentioned in another thread, resolution will never be worse as a consequence of greater pixel count, whatever the aperture, but this might not be true if one is comparing different systems, such as the Olympus E-3 with the Canon 40D.

[Panopeeper]

System resolution is generally a combination of lens resolution and film or sensor resolution. If the resolution of either one is increased, whilst keeping the other the same, the result will likely (but not necessarily) still be a modest increase in system resolution, depending on other factors such as the strength of the AA filter

Great. Now, we have a case, where the sensor resolution is increased, while the lens resolution decreases due to diffraction.

This is apart from the fact, that the formula you used in your calculation is though widely circulated, but useless, i.e. not applicable. Nothing, but really nothing can make up for the loss caused by the lens.

[Ray]

Of course, the larger/more effective microlenses increase the amount of captured light. This leads to less noise compared to a shot with teh same exposure but less effective microlenses. However, again, this leads to earlier saturation as well.
[a href=\"index.php?act=findpost&pid=217467\"][{POST_SNAPBACK}][/a]

Gabor,
I think this is another example of your not seeing the woods for the trees. You seem to be too concerned with the performance of the individual pixel or sentel at the expense of the over all image. What's important with regard to the final result, is the performance per unit area of sensor.

At the pixel level, each one of 4 small sentels, taken individually, are almost certainly going to be noisier and have less DR than one big pixel covering the same area as the 4 smaller pixels. However, from the perspective of the image as a whole, we're interested in how the 4 smaller pixels (as a group) compare with the one larger pixel covering the same area.

[Panopeeper]

At the pixel level, each one of 4 small sentels, taken individually, are almost certainly going to be noisier and have less DR than one big pixel covering the same area as the 4 smaller pixels. However, from the perspective of the image as a whole, we're interested in how the 4 smaller pixels (as a group) compare with the one larger pixel covering the same area.

Ray,

I am not participating in your attempts to convert a camera into another one. If I buy a 15 Mpix camera, then I don't compare that with a 2 Mpix, because I don't want to downres the images. If I wanted to, then I would buy another camera instead of the 15 mpix.

In other words: four pixels are not one pixel.

(I do not intent to buy the 50D; I am with Bernard: I need more DR, not more pixels. And no, I am not interested in converting the DR into something else.)

[Ray]

Great. Now, we have a case, where the sensor resolution is increased, while the lens resolution decreases due to diffraction.

This is apart from the fact, that the formula you used in your calculation is though widely circulated, but useless, i.e. not applicable. Nothing, but really nothing can make up for the loss caused by the lens.
[a href=\"index.php?act=findpost&pid=217473\"][{POST_SNAPBACK}][/a]

Any decrease in lens resolution has nothing to do with pixel count. Lens resolution is dependent on lens quality and aperture selected.

If we are comparing same size sensors, then choice of aperture will be very closely the same for same DoF whatever the pixel count, but not exactly the same.

If we are comparing different size sensors, then choice of aperture will be different for same DoF in accordance with the crop factor.

[Tony Beach]

...resolution will never be worse as a consequence of greater pixel count, whatever the aperture[a href=\"index.php?act=findpost&pid=217469\"][{POST_SNAPBACK}][/a]

No, but resolution won't be any better when you hit the diffraction wall.  Lots of consumers are going to be scratching their heads when they don't see any improvement going from a 40D to a 50D shooting at f/11.

[Ray]

Ray,

I am not participating in your attempts to convert a camera into another one. If I buy a 15 Mpix camera, then I don't compare that with a 2 Mpix, because I don't want to downres the images. If I wanted to, then I would buy another camera instead of the 15 mpix.

In other words: four pixels are not one pixel.

(I do not intent to buy the 50D; I am with Bernard: I need more DR, not more pixels. And no, I am not interested in converting the DR into something else.)
[a href=\"index.php?act=findpost&pid=217477\"][{POST_SNAPBACK}][/a]

Gabor,
You seem to have misunderstood my analogies. I'm comparing a large pixel with 4 small pixels (or a 3mp sensor with a 12mp sensor) for the sake of simplicity and in order to highlight an underlying principle.

In practice, these incremental increases in pixel count from Canon are like comparing one small pixel with a previous one just 1.2x the size. But the principle is the same. If you apply a bit of new technology, such as reducing the gap between microlenses, reducing the effect of extraneous noise within the system, reducing the size of on-chip processing devices (whatever they may be), then you can end up with a camera capable of producing improved image quality.

[Panopeeper]

Any decrease in lens resolution has nothing to do with pixel count. Lens resolution is dependent on lens quality and aperture selected

Lens resolution has to do a lot with pixel resolution (not with pixel count), namely if the lens does not resolve enough, then the high pixel resolution is useless.

I'm comparing a large pixel with 4 small pixels (or a 3mp sensor with a 12mp sensor) for the sake of simplicity and in order to highlight an underlying principle

I find these comparisons irrelevant and your underlying principles misguided, just like your mixing up the circle of confusion with sensor size.

However, I do not intend to convince you from the correctness of my viewpoint.

[BernardLanguillier]

There is no such thing. The sensors' output is strictly linear (disturbed somewhat by the noise).
[a href=\"index.php?act=findpost&pid=217465\"][{POST_SNAPBACK}][/a]

My experience with previous DSLR (starting with the D2x) clear shows that the sensor is not linear in the .1 brighest stop where it matter most. This wouldn't show in your test that does not focus on very bright illuminations like those found at the edge of the sun for instance.

The D3 does improve a lot on this.

Cheers,
Bernard