• Please use real names.

    Greetings to all who have registered to OPF and those guests taking a look around. Please use real names. Registrations with fictitious names will not be processed. REAL NAMES ONLY will be processed

    Firstname Lastname

    Register

    We are a courteous and supportive community. No need to hide behind an alia. If you have a genuine need for privacy/secrecy then let me know!
  • Welcome to the new site. Here's a thread about the update where you can post your feedback, ask questions or spot those nasty bugs!

Deep Shadows to Bright brights: taming the span of light!

John Sheehy

New member
Well, it turns out that Canon's 1DMKIII has 7.2micron pixels, which seem "smaller" than the N's 8.2microns, right? Well...Wrong!

Stop guessing what I might say about something. So far, you haven't followed what I have actually said, no mind what I might say.

Canon cleverly managed to reduce on-board circuitry "footprint", around the "sensels", while increasing size of micro-lenses (even better fill-ration) and still keeping the *same* surface/size of the photo-sensitive element in the sensel.

It's about time. The 1DmkII was very inefficient, collecting approximately the same number of photons per pixel as the 20D/30D at the same ISO, with a larger pixel pitch. The 1DmkIII is playing catch-up to 20D efficiency.

WOW! The cramped two additional million pixels in the wonderful APS-H 1.25x sensor, while *potentially* (my opinion) increasing quantum efficiciency due to better fill-factor and same photo-surface.

That's a step forward, but overdue, IMO. Why did they waste so many photons in the mkII?

Canon is surely sensitive to the mathematics implied here, and seems to be far away from anything that you are proposing, in terms of smaller (and worse) pixels. They are avoiding that and for very good reasons...

They are avoiding it for any number of non-IQ related reasons, such as card write speed, storage concerns, and pixel peepers who don't understand that the pixel has a variable influence on the image, based on displayed size.

And Roger is right on the money, too.

About what? You seem to be purposely engaging in vague debate and hiding behind authority, and drama like "WOW!", BINGO", and "Right On!.

Roger is correct, that smaller pixels are noisier instruments. He is wrong in suggesting that image quality is directly related to pixel quality. I have demostrated otherwise. The 10D image, with it lower pixel noise, looks just as noisy as the FZ50 pixels when blown up to the same subject magnification. You can't ignore that fact. Displayed size of original pixels is every bit as important as standard deviations, a vital concept that he does not address.
 

Asher Kelman

OPF Owner/Editor-in-Chief
Roger is correct, that smaller pixels are noisier instruments. He is wrong in suggesting that image quality is directly related to pixel quality. I have demostrated otherwise. The 10D image, with it lower pixel noise, looks just as noisy as the FZ50 pixels when blown up to the same subject magnification. You can't ignore that fact. Displayed size of original pixels is every bit as important as standard deviations, a vital concept that he does not address.
Hi John,

I can't argue for or against your "displayed size argument" using your route of pixel manipulation, since I am having difficulty finding a practical example for my work. Let's consider a this route, if I may.

Test: Picture of a painting: I want the best quality image.

The end point: The least noise/unit area on the final print.

1. Picture with the 30D or the FZ50.

I'd use whatever focal length in each case to get the best image focused on the entire camera sensor.

Now could you reframe your argument in that practical situation.

Here the digicam file would have to be blown up more. So why wouldn't that image degrade even more that that of the 30D to get to make a final print of say 16" x 20"?

There are less, smaller and noidier pixels in the FZ50 that has to be upressed much more which should make the picture have magnified noise.

By contrast the 30D image would start and end with less noise per unit area.

So in this practical example, why can't we say that smaller pixels (in the Digicam) are noisier in the final magnified image.

2. Equal Size Camera Sensor Chips:

Now if we would start with identical size CCD /Pixel sensor chips, with the only difference being pixel size, for sure there is more noise/unit area with many small noisy pixels (although Moiré may be prevented).

So tell me where your concepts fit in in my practical example taking a photograph of a picture and filling the entire sensor chip surface and not wasting any pixels?

Asher
 
Last edited:

Ferenc Harmat

New member
More factual data, here...

It's about time. The 1DmkII was very inefficient, collecting approximately the same number of photons per pixel as the 20D/30D at the same ISO, with a larger pixel pitch.

That's pretty weird, because, being myself the owner of both cams, I did not find that to be true. Just because both cams collect close to 51K photons @ ISO100 does not mean anything about the true or max. gains you can realize, relatively speaking.

Not only that, I have posted a sample that shows pretty much the contrary. Additionally, the 1D MarkII/N collects about 80K photons at ISO75-80, which is about 0.30EV below ISO100 on the 30D (not counting 30D's extra +0.25EV of "sensitivity" over ISO100), which collects around 51K. And it shows: our 1D MarkII-N keeps going where our 30D ends.


That's a step forward, but overdue, IMO. Why did they waste so many photons in the mkII?

Well, fact is they are not wasted. I already demonstrated it.

You seem to be purposely engaging in vague debate and hiding behind authority, and drama like "WOW!", BINGO", and "Right On!.

John, the problem is that you have not been able to clearly and fully support your arguments with the necessary evidence, other than a general description of the issue. I chose, on the other hand, to resort to universally available sources that are, indeed, far better references than a simple personal opinion (including mine).

That, I believe, has little or nothing to do with "hiding" or anything like that. You are encouraged, as well, to give more consistency or validity to your points. That is all.


He is wrong in suggesting that image quality is directly related to pixel quality. I have demostrated otherwise.

He is right, because when you work in Astro-imaging, *every single* pixel counts for anything you do, for anything you see. That is where Roger is coming from, which, by the way, is an application that pretty much challenges any sensor and technology out-there.

It is, by far, a much more compelling background and show-case for what today sensors can or can not do, especially when compared to the image of a puppet that does not allow to draw any conclusion other than more pixel density means more resolution.

Again, please feel free to expand our view with a similarly compelling case or application, so I can see your points in motion, more clearly.

At this time, the results are clear and evident to me. And Canon seems to be working in the exact same direction.

Just my 0.02.
 

John Sheehy

New member
Hi John,

I can't argue for or against your "displayed size argument" using your route of pixel manipulation, since I am having difficulty finding a practical example for my work. Let's consider a this route, if I may.

There is nothing practical available. I am trying to demonstrate what a 47MP 1.6x-crop camera could be like, compared to the 10D, using Panasonic's tiny pixels and current readout technology.

Test: Picture of a painting: I want the best quality image.

The end point: The least noise/unit area on the final print.

1. Picture with the 30D or the FZ50.

I'd use whatever focal length in each case to get the best image focused on the entire camera sensor.

Now could you reframe your argument in that practical situation.

My argument has no relation to that situation, except to say that if there were a camera that did a full 36*24mm of FZ50 pixels, and you had a lens to match, that would give a better image than something like a FF Canon does now, at the same ISO, especially at ISO 100.

Here the digicam file would have to be blown up more. So why wouldn't that image degrade even more that that of the 30D to get to make a final print of say 16" x 20"?

At small sizes and ISO 100, I don't think you'd see a huge difference, but at some point of enlargement, the digicam will start showing its higher noise, due to its capture of a lot less photons per unit of displayed area. The difference should be easily visible at that size (16x20), especially for the higher ISOs.

There are less, smaller and noidier pixels in the FZ50 that has to be upressed much more which should make the picture have magnified noise.

By contrast the 30D image would start and end with less noise per unit area.

Yes, as long as you are talking about displayed area, in which case the FZ50 physical captured image has to be magnified much more than the 30D's.

If you were talking about noise per unit of sensor area, then the FZ50 and the 30D have similar shot noise per unit of area, but the FZ50 has less read noise per unit of area at ISO 100.

So in this practical example, why can't we say that smaller pixels (in the Digicam) are noisier in the final magnified image.

We can say that; I've never implied otherwise.

2. Equal Size Camera Sensor Chips:

Now if we would start with identical size CCD /Pixel sensor chips, with the only difference being pixel size, for sure there is more noise/unit area with many small noisy pixels (although Moiré may be prevented).

My point is that this is false. There is no noise benefit per unit of sensor area with larger pixels, per se; only with better effective fill factors. There is a small benefit in many real world applications, only because of the way manufacturers butcher their RAW data. The ISO 1600 RAW data from the FZ50 is far worse than it needs to be. Panasonic uses a low-quality amplifier, that adds significant noise to the signal at high ISOs - I'm not talking here about amplifying unavoidable sensor noises (dark current and shot); I'm talking about creating new noise in the readout chain). Under-exposing 4 stops at ISO 100 gives better results for RAW (but not JPEG) than using ISO 1600. Like most manufacturers, they also clip their RAW data at the blackpoint, which guarantees bright, noisy blacks. Canon doesn't do this to their RAW data (although they may commit other RAW purity crimes). The optimal way to do demosaicing and white-balance is to do it in the linear state of the RAW data, and in that state, leaving a blackpoint bias in the RAW data does not interfere with white-balancing and demosaicing, and the act of scaling the color channels into negative space as well as into positive space balances them towards black better, and the same for demosaicing, and even moreso if you need to do any downsampling - all the color bias and intensity of noise diminishes if these "negative" values remain (and they also contain a small amount of signal, as well). A 120MP full-frame camera could be RAW-converted in such a way that the darkest areas are rendered from a lower-resolution version, downsampled while still containing negative noise for real blacks, and the midtones and highlight with a full-res, hi-Q demosaicing.

None of these issues have any connection to pixel size and shot noise. The *necessary* shot noise of these small pixels is only part of the noise you see (none of the line banding, for instance, has anything to do with pixel size).

So tell me where your concepts fit in in my practical example taking a photograph of a picture and filling the entire sensor chip surface and not wasting any pixels?

It's not about that at all. I don't recall implying in any way that an FZ50 could ever give full image noise better than a 30D, or any other DSLR. What I am trying to show is what it would be like to have a 47MP 1.6x-crop DSLR, with *REAL* tiny pixels, from a real camera. And I'm sure Canon can do a better job than Panasonic with the read noise, and Canon leaves the negative noise in their RAW data.

Shot noise is only an enemy for images from a camera with a small total sensing area (sensor size times effective fill factor, all other things being equal); not small pixels per se.
 

John Sheehy

New member
That's pretty weird, because, being myself the owner of both cams, I did not find that to be true. Just because both cams collect close to 51K photons @ ISO100 does not mean anything about the true or max. gains you can realize, relatively speaking.

How does it not mean that? You're not making any sense - shot noise is determined by photon collection. Canon brags in their white papers about how much their big-pixel 1.3 and FF cameras combat noise at all ISOs with their larger photon collection, yet their existing bigger-pixel cameras don't collect any more photons per pixel than the 20D with the same illumination on the sensor. My take is that they saw enough improvement just because of the way that these cameras are less demanding of the lenses with the bigger pixel pitch, and didn't accurately assess what is the real benefit.

The 1-series has lower read noise at the pixel level than all the other Canon DSLRs at ISOs 100 and 200, but that has nothing to do with photon collection. The Pentax K10D, an APS-sized 10MP has lower read noise at ISO 100 than the 1-series Canons.

Not only that, I have posted a sample that shows pretty much the contrary. Additionally, the 1D MarkII/N collects about 80K photons at ISO75-80, which is about 0.30EV below ISO100 on the 30D (not counting 30D's extra +0.25EV of "sensitivity" over ISO100), which collects around 51K. And it shows: our 1D MarkII-N keeps going where our 30D ends.

You are one of the few people I've seen raving about Canon's ISO 50 on the DSLRs. Most claim that DR is limited.

The read noise on recent Canon DSLRs drops significantly going from ISO 1600 to 800 to 400, and then level off at a minimum (possibly dictated by the ADC circuitry). The ISO 50 should have the same read-noise limit to shadow quality, plus it doesn't collect enough photons to get a full range of highlights. ISO 50 should only work much better than 100 for low-contrast or high-key subjects, and then you have to work to avoid clipping.

The camera really should have been ISO 64 as a base ISO; a real ISO at the non-standard value would create a minimum ISO without compromise. Just one of many bad decisions made by Canon.

Well, fact is they are not wasted. I already demonstrated it.

Nonsense, you just admitted that the 1DmkII collects about the same number of photons at the same ISO as the 20D, with a pixel coverage area 1.69x as large. This means that the 20D is 1.69x as efficient. Photons are turning into heat or reflecting back out through the lens in 1dMKII. That is inefficient. Canon is misrepresenting the facts about why the 1-series and 5D cameras can give bigger images than their APS DSLRs. The benefit is purely the sheer number of pixels for the 1DsmkII (and the 5d), and the lower demand made on the lenses by both 1-series cameras and the 5D. And, of course, lower read noise on the 1-series at ISOs 100and 200 (which is due to electronics, mainly).

John, the problem is that you have not been able to clearly and fully support your arguments with the necessary evidence, other than a general description of the issue. I chose, on the other hand, to resort to universally available sources that are, indeed, far better references than a simple personal opinion (including mine).

Did you even look closely at the cat pictures? The 10D, with its larger pixels and lower noise at the pixel level, is just as noisy in the real world with the same ratio of displayed size to sensor crop size. This illustrates that noise as determined by standard deviations, is only relevant for viewing pixels at a certain size in display. Standard deviation is only one parameter in *image* noise (even at the same pixel display size, visible noise is not directly related to standard deviation - there are different shapes and spectrums of noise). Take a 12x8 pixel crop of blue sky at a low ISO on your mkII, and upsample it to 1200x800 - all of a sudden, 100x100 tiles appear, with slight differences in color. You only failed to see the noise at normal size because the noise was small - the standard deviation hasn't changed.

Did you do what I suggested with Bart's water-tower images? Downsample a copy to 33%,and give it 1/3 the noise of the original, and then bring it back up to size?

The larger pixel does not win, not even in visible noise. Open your mind and open your eyes! This pixel-centric noise obsession is a disease, leading people to all kinds of false conclusions.

That, I believe, has little or nothing to do with "hiding" or anything like that. You are encouraged, as well, to give more consistency or validity to your points. That is all.

Well, you go off on a tangent with experiments that compare large sensors to small ones. How did you think that Roger's comparison of the 1DmkII and the S70 had anything to do with what I was talking about?

He is right, because when you work in Astro-imaging, *every single* pixel counts for anything you do, for anything you see. That is where Roger is coming from, which, by the way, is an application that pretty much challenges any sensor and technology out-there.

There are no cameras out there that have 47MP APS sensors or 120MP 36x24mm sensors; none that regular people can buy, anyway. That's why I have to do a simulation, and it really isn't much of a stretch: we just can't see the whole image.

It is, by far, a much more compelling background and show-case for what today sensors can or can not do, especially when compared to the image of a puppet that does not allow to draw any conclusion other than more pixel density means more resolution.

No, the most important point is that the larger-pixel camera has just as much noise, when the actual focal plane crops are magnified the same amount. Most analyses are concentrating on the Z-axis, and *TOTALLY* ignoring the X and Y factors of noise.

Again, please feel free to expand our view with a similarly compelling case or application, so I can see your points in motion, more clearly.

At this time, the results are clear and evident to me. And Canon seems to be working in the exact same direction.

What is "clear and evident"? How many experiments like mine have you seen? Most everyone, Roger included, is testing the wrong things if you would like to know what it would be like to have 9 pixels in the space of today's DSLR's one. Comparing similar FOVs from an S70 and a 1DmkII will never help with that.
 
Last edited:
There is no noise benefit per unit of sensor area with larger pixels, per se; only with better effective fill factors.

Even with the 'per se' nuance, be careful!

A larger photon collecting surface will, everything else being equal, collect a proportionally larger number of photons per sensel in the same timeframe. That by itself will improve photon shot noise, which equals the square root of the number of photons collected by each sensel. Therefore, each doubling of linear sampling density (or 1/2 sensel pitch) will (slightly more than) double the shot noise.

A proportionally larger well area will fill just as fast, due to the larger number of photons being collected per unit time.

In addition, when larger numbers of electrons are available, it also makes more sense to use higher quality (14-16 bit) ADCs. For a very small sensel with a saturation level in the order of 10000 electrons, one cannot reasonably expect much (if any) benefit from a >12-bit ADC when we assume 1 bit of other (read and thermal) noise, especially when pushing the ISO above the unity gain level.
A larger capacity well, say in the 50000 electron range, would be best served with a 14+ bit ADC, also assuming 1 bit of useless noise, which is exactly what the latest Canon EOS-1 offering suggests to be achieving.

Bart
 
Last edited:
- shot noise is determined by photon collection. Canon brags in their white papers about how much their big-pixel 1.3 and FF cameras combat noise at all ISOs with their larger photon collection, yet their existing bigger-pixel cameras don't collect any more photons per pixel than the 20D with the same illumination on the sensor.

Unfortunately this is formulated in an easy to misinterpret manner, to me anyway. It is unfortunate because the ambiguity raises more questions about, than it clarifies, your statement.

I'm a supporter of a less ambiguous terminology.
A sensor is generally understood as the entire sensor array device, I therefore prefer sensor array.
A sensor element (sensel) is the smallest uniform part by which the array is filled.
A pixel represents the smallest unique part of an output image, in this context the result of demosaicing and postprocessing multiple quantized R/G/B sensel values. There is of course no law forcing the adoption of this terminology just because I say so, but it would follow the ISO terminology more closely.

The 'sensor' at the end of the above quote, could be misinterpreted as sensor element, because of your earlier use of 'pixel' in the same sentence. Perhaps you spoke about the same integrated amount of light on different size sensor arrays, but that sofar remains a guess on my part.

I (as a not native speaker of the English language) wouldn't dare to criticize native speakers on their choice of words. I'm just feeding back the confusion it can cause, and as a result confuse the discussion, which I assume is not intended.

Peace,
Bart
 

Ferenc Harmat

New member
Not correct / accurate, *at all*...

You are one of the few people I've seen raving about Canon's ISO 50 on the DSLRs. Most claim that DR is limited.

That is precisely the difference between just reading and... finding the truth.


plus it doesn't collect enough photons to get a full range of highlights. ISO 50 should only work much better than 100 for low-contrast or high-key subjects, and then you have to work to avoid clipping.

Again, the difference between a "supposed" versus an "observed" event could be definitely significant.

Your comment seems to be weakly related to what ISO50 means on 1D-class bodies (at least on our 1D MarkII-N body). On this body, ISO50 means ISO80, and at this level, it has clearly SUPERIOR signal-to-noise ratio and I am amost 100% sure it does offer better dynamic range, although the improvement is modest (but there IS and improvement).

The evidence speaks for itslef. On the samples I posted (*my* 30D vs. *my* 1D MarkII-N), the N is simply superior, at ISO80. In fact, it is sensibly superior, especially in the shadows and overall high-frequency retention.

Please, provide feel free to provide some evidence or images that can prove the opposite and/or back-up your comments. It will really help. Hopefully, you also own, operate and demand from both cams the best they give, either for pleasure or work. I am not surprised that I am among the few that are currently or recently exploiting this... although I heard about this almost a year ago but never paid attention. If no attention to detail is really put in the subject, no one will ever know.

Nonsense, you just admitted that the 1DmkII collects about the same number of photons at the same ISO as the 20D, with a pixel coverage area 1.69x as large.

John, I intentionally gave you the chance (and even clues) on my comment to allow you to re-think your statement. But it seems that you totally missed them or simply forgot that ISO100 on the 30D *is not* ISO100. It is, in fact, around +0.3EV of effective sensitivity (ISO125), but the catch here is that this sensitivity *is not* the product of increased quantum efficiency or anything like that.

Evidence suggests that it is just mostly pre-set gain, "volume raised a notch" at Canon's design labs. This also explains why on this world, my 30D is *cleary and consistently* noiser on the shadows at basically any ISO speed (from ISO100) when compared to my N.

This ultimately explains why the N is capable of putting such an amazing performance at 30secs (ISO80) and clearly beating my 30D on shadow detail (global) and high-frequency response and retention Micro-detail in midtones-to-shadows) at 20secs... YES! It should have been able to reach final exposure at around 10-15 secs (assuming ISO125), yet it could not handle the shadows properly until reaching 20secs, and the N, with just 10 more secs. at ISO80 just blew it away (more shadow detail, highlights well conserved and much, much less "blobs" of chroma noise in all channels). I am happy camper, and this can be certainly used at ANY time. Simply set your cam at "L" (1D MarkII/N) and dial a permanent -2/3EV. Now, go out and take your pictures (day or night) and enjoy higher-signal to noise ratio, especially visible in deep shadows).

Why all this? BIGGER and better sensels, potentially allow you to produce better sensor-arrays, which in turn, potentially allow you to extract better pixels and better images. Now, how you extract them is a different topic, altogegther.



No, the most important point is that the larger-pixel camera has just as much noise, when the actual focal plane crops are magnified the same amount. Most analyses are concentrating on the Z-axis, and *TOTALLY* ignoring the X and Y factors of noise.

Since you seem so clearly focused and convinced of this, I am all ears. Everything you can share, show and tell to further expand this point, I will seriously look at and consider as part of the discussion.
 
Top