larger sensors

[eronald]

Establish a ttrace length on the enlarged print that corresponds to the max allowable shake. Then bactrack from there ?

Edmund

The question for me is, having established an adequate shutter speed for a tack sharp image with a given lens and format, how much faster does that shutter speed need to be if we increase the pixel count, and print size in proportion, but keep the format and lens the same?

My view is, we should increase the shutter speed in proportion to the increased resolving power of the sensor, but not for equal size prints, but for prints that express the native resolution of both sensors, at a ppi sufficient to convey the maximum detail to the paper, that can be detected by anyone with normal vision from a 'reading' distance.
[a href=\"index.php?act=findpost&pid=93352\"][{POST_SNAPBACK}][/a]

[bjanes]

Edmund,

I should add, for the benefit of Bill Janes, that the question as to whether or not the 1/FL rule is adequate for sharp images is quite irrelevant. I use it purely for illustrative purposes. One has to have a reference point. Make it 1/4FL if you like.

The question for me is, having established an adequate shutter speed for a tack sharp image with a given lens and format, how much faster does that shutter speed need to be if we increase the pixel count, and print size in proportion, but keep the format and lens the same?

My view is, we should increase the shutter speed in proportion to the increased resolving power of the sensor, but not for equal size prints, but for prints that express the native resolution of both sensors, at a ppi sufficient to convey the maximum detail to the paper, that can be detected by anyone with normal vision from a 'reading' distance.
[{POST_SNAPBACK}][/a]

I would agree with Ray here. However, as pointed out in the Stanford article, the ability to increase the shutter speed may require higher ISO and this may be limited by small pixels. This is usually not a problem with daylight, but it is with available light exposure.

BTW, the linear blur discussed in the article can be reduced with Photoshop's smart sharpen filter, but the random walk type of blur with longer exposure times is more problematic.

There is a trade off between pixel size dynamic range and noise, especially at high ISO. Ray's 25 MP 5 um pixel full frame camera would be great for landscapes but not so good for available light work. Presently, according to most surveys, most 35 mm type digital users are more interested in low noise, high DR, and good high ISO performance. Canon will introduce Ray's camera they can control the noise, IMHO. I still maintain that the use of a tripod would be necessary to take advantage of the increased resolution, but a tripod would not be necessary to have the same detail as with a lower MP camera.

[a href=\"http://www.clarkvision.com/imagedetail//does.pixel.size.matter2/]Roger Clark[/url] has posted an addendum to his post on pixel size. It demonstrates some of these topics with pictures of a night scene.

Bill

[Ray]

There is a trade off between pixel size dynamic range and noise, especially at high ISO. Ray's 25 MP 5 um pixel full frame camera would be great for landscapes but not so good for available light work. [a href=\"index.php?act=findpost&pid=93362\"][{POST_SNAPBACK}][/a]

I don't see it, Bill. It's almost as though you are saying all technological development with regard to improved image quality has come to an end and that more pixels just means more read-noise.

We know with the 400D (if Canon is to be believed) that they reduced the gap between microlenses, which results in a greater amount of light reaching the individual photodiodes than would otherwise take place. It's not clear if the 400D microlenses are the same size as the 30D microlenses, as a result of reducing that gap, or just closer to that size than they otherwise would be.

Nor is it clear if the actual photodiodes themselves are the same size or smaller. As you know, pixel pitch is always considerably larger than photodiode size on a CMOS sensor. The first Canon DSLR, the D30, had a pixel pitch of around 10 microns, but a photodiode size of only 5.25 microns. The rest of the space was presumably taken up with on-chip processors.

Unless you are really 'in the know', or a research scientist at one of Canon's laboratories, it cannot be clear what improvements are potentially there to be made. When Canon announced they had reduced the microlens gap in the 400D, I was surprised because I had assumed the gap was already as small as it could be.

For all I know, when Canon introduce their 24, 25 or 22mp FF 35mm sensor, they might also claim an improved dynamic range over previous models, at ISO 50, due to increasing the actual size of the photodiode and reducing the size of the on-chip processors, or sticking the on-chip processors on the other side of the chip, or creating a separate chip for all, or some of the processing.

Of course, it almost goes without saying, if you want increased dynamic range in a system that is mostly limited by photonic noise, it has to be through increased exposure. You can't have increased dynamic range as well as faster shutter speeds. No matter how many pixels are on the sensor, the sensor as a whole receives the same amount of light for a given exposure at a given f stop.

I'll have to edit this in case someone tries to argue that a 200mm lens at f8 lets more light pass for a given exposure than a 50mm lens at f8   . I am of course referring to a situation of equal size sensors, ie. equal formats.

[bjanes]

I don't see it, Bill. It's almost as though you are saying all technological development with regard to improved image quality has come to an end and that more pixels just means more read-noise.

[{POST_SNAPBACK}][/a]

The physics of CMOS and CCD is very well understood. Improvements will be made, but physical limits are being approached. Current Canon sensors have read noise of about 3-4 electrons and microlens technology is relatively advanced. The Poisson distribution still applies to photon sampling. Fill factors are relatively high. There could be a two fold or greater improvement in quantum efficiency. Finally, the improvements made for small pixels will also apply to large pixels. Pixel size does matter.

Large and small pixels have similar read noise, but the effect on the image is much greater with small pixels because the large pixel accumulates many more electrons (i.e. the large pixel has greater gain).

Why don't you read Roger Clark's article?

[a href=\"http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/]http://www.clarkvision.com/imagedetail/dig...rmance.summary/[/url]

[Ray]

The physics of CMOS and CCD is very well understood.

It is indeed, Bill. But by whom? That's the question? Certainly not by me, and without intending to offend you, probably not by you, or Roger Clark.

Since I first started reading LL, and other similar forums, I've come across a littany of so-called experts claiming that fewer big pixels are better than more small pixels. (There must be a grammatically neat way of expressing that   ).

They've mostly proved to be wrong, with the passage of time. A 10D pixel, a 20D pixel and a 1Ds2 pixel is better than a 1Ds, D60 or D30 pixel, from the point of view of all the things that count with regard to image quality. Is this not true?

[Ray]

Current Canon sensors have read noise of about 3-4 electrons and microlens technology is relatively advanced. [a href=\"index.php?act=findpost&pid=93399\"][{POST_SNAPBACK}][/a]

A quick calculation tells me a 35mm sensor with a pixel pitch of 5.25 microns contains around 30mp. Microlenses may not be required, Bill.

How does this grab you? No microlens. No AA filter. On one side of the chip nothing but wall to wall photodiodes 5 microns in dimaeter. On the reverse side of the chip, nothing but analog amplifiers, A/D converters and processors of various types and function. Somewhere else, the most powerful computer and processing algorithms to date, in a Canon camera.

[eronald]

Note the way the pixels work is changing - on the recent Canons, pixel values are the delta between the start and end of exposure.

As to who knows about this ? I guess any of us engineers could go to an imaging systems design conference and get the latest info. All the companies are eager to publish some details of the chips they put out, once the design is finished, if only as a recruiting tool.

Edmund

It is indeed, Bill. But by whom? That's the question? Certainly not by me, and without intending to offend you, probably not by you, or Roger Clark.

Since I first started reading LL, and other similar forums, I've come across a littany of so-called experts claiming that fewer big pixels are better than more small pixels. (There must be a grammatically neat way of expressing that   ).

They've mostly proved to be wrong, with the passage of time. A 10D pixel, a 20D pixel and a 1Ds2 pixel is better than a 1Ds, D60 or D30 pixel, from the point of view of all the things that count with regard to image quality. Is this not true?
[a href=\"index.php?act=findpost&pid=93403\"][{POST_SNAPBACK}][/a]

[BJL]

Large and small pixels have similar read noise, but the effect on the image is much greater with small pixels because the large pixel accumulates many more electrons (i.e. the large pixel has greater gain).
[a href=\"index.php?act=findpost&pid=93399\"][{POST_SNAPBACK}][/a]

You assume that sensor dynamic range will always be limited by a maximum recordable signal (photo-electron count), limited in turn by maximum well capacity, with this limit higher with photo-sites of larger area. Maybe so, but this is far from certain for all possible sensor technologies, which might be able to make maximum photo-electron counts irrelevant beyond a quite modest photo-site size.

One current possibility is technologies like Fujifilm's SuperCCD SR, expanding highlight headroom beyond the standard well capacity limits (of about 1000e per square micron). Yes, larger sensors can expand it even more, but the law of diminishing returns sets in once a given photo-site size gives more than enough DR for almost any situation.

Another possibility is that a future sensor could read and reset electron wells multiple times during an exposure, eliminating any maximum well capacity. Perhaps some version of progressive scan as used for video, but at far higher than normal video frame-rates.

Another idea, better for keeping shadow noise levels down, is already implemented in sensors for video surveillance cameras. Roughly, a sensor could allow highlight photo-sites to fill up, measure the time taken for such photo-sites to fill, and extrapolate to the photo-electron count that would have occurred over the full exposure time.
If this full well count is at least as high as the signal given by mid-tones in the Canon 5D at ISO 100, about 5,000 electrons, the S/N ratio at such a photo-site is dominated by photon shot noise and is about 70:1 or better. This is easily enough to avoid visible noise, as indicated by the lack of noise problems in mid-tones from the 5D at ISO 100. At a rough estimate, 5,000e capacity is possible with pixel pitch of under 2.7 microns. (The 5.4 micron photo-sites of the Olympus E-500 sensor have capacity 25,000e, so 2.7 microns should be capable of 1/4 that.) That pixel size would be enough for over 50MP in DX format, and probably enough to put the resolution limits entirely on SLR lenses

I believe that the surveillance video sensors repeatedly check each photo-site to see if it is close to full at 1/2, 1/4, 1/8 etc. of the full exposure time: near full photo-sites are read and A/D converted right at the photo-site, adjusting for the reduced exposure time by bit shifts.


Such approaches could essentially allow arbitrarily low ISO speeds for good shadow handling. That leaves shutter speed limits determined mainly by maximum usable aperture diameter, which is indirectly related to format size through optical limitations on how low aperture ratios can be.

[Morgan_Moore]

A car is traveling at one meter per second

An exposure is made of one second

The car has one meter of blur

Irrelevant of capture method / size of sensor / lenght of thread etc

With a higher resolution you will be able to appreciate the blur in greater detail

Tell me I am wrong ?    

[bjanes]

It is indeed, Bill. But by whom? That's the question? Certainly not by me, and without intending to offend you, probably not by you, or Roger Clark.

Since I first started reading LL, and other similar forums, I've come across a littany of so-called experts claiming that fewer big pixels are better than more small pixels. (There must be a grammatically neat way of expressing that   ).

They've mostly proved to be wrong, with the passage of time. A 10D pixel, a 20D pixel and a 1Ds2 pixel is better than a 1Ds, D60 or D30 pixel, from the point of view of all the things that count with regard to image quality. Is this not true?
[{POST_SNAPBACK}][/a]

No offense taken, Ray, since I do not claim to be an expert. At least I have the sense to listen to those who know more than I do. Roger Clark is an imaging expert, having received his PhD in planetary science from MIT and having worked on NASA imaging probes. In addition, he has published 179 peer reviewed scientific papers.

[a href=\"http://www.clarkvision.com/rnc/index.html]Bio Roger N Clark[/url]

Also, the illustrious host of this site has written a good essay on this subject. He is not an engineer, but he does have the ear of many experts from Thomas Knoll to the engineers at the Dalsa chip FAB, which he recently visited.

Micheal's Sense & Sensor

In the other corner, we have Ray.      

Bill

[Ray]

In the other corner, we have Ray.      
[a href=\"index.php?act=findpost&pid=93523\"][{POST_SNAPBACK}][/a]

I understand what you're implying, Bill. Accepting that someone is right because he has a stack of publications behind his name, a few awards and perhaps a PhD, is very easy. It's something that appeals to our inbred respect for authority.

However, as an Australian, I don't have much regard for authority   . There's a tradition here of throwing off the shackles of our imperial masters.

It seems I take a more philosophical view of these matters than you. The most brilliant experts in their field can often be wrong (and invariable are proved to be wrong, in the fullness of time, on at least a few issues). Even Eistein appears to have been wrong on a few counts.

We're talking here about cutting edge matters in a very specialised branch of physics. Developments at the coal face are no doubt the subject of NDAs and neither Roger Clark nor Michael may be privy to them, for all I know. At least I wouldn't too readily assume that they are.

On matters of what's possible through the application of the scientific method, I'm an optimist. I get the impression, Bill, that you are more concerned with why things can't be done.

[howiesmith]

On matters of what's possible through the application of the scientific method, I'm an optimist.

[a href=\"index.php?act=findpost&pid=93534\"][{POST_SNAPBACK}][/a]

I have found the difference between optimists and pessimists is pessimists have better data.

Ray, are you saying you have a good chance of being right because you are ignorant?

[Ray]

Ray, are you saying you have a good chance of being right because you are ignorant?
[a href=\"index.php?act=findpost&pid=93536\"][{POST_SNAPBACK}][/a]

No, I'm saying (or at least implying) that we are all ultimately ignorant, but some more than others. What is perhaps more important is knowing that you don't know rather than believing that you do know.

Of course it can get a bit complicated because it's possible to to believe that you don't know when in fact you do know. You just didn't know that you knew.  

[howiesmith]

No, I'm saying (or at least implying) that we are all ultimately ignorant, but some more than others. What is perhaps more important is knowing that you don't know rather than believing that you do know.

Of course it can get a bit complicated because it's possible to to believe that you don't know when in fact you do know. You just didn't know that you knew. 
[a href=\"index.php?act=findpost&pid=93540\"][{POST_SNAPBACK}][/a]

So, are you saying that the less one knows (more ignorant), and the less one knows he knows (the more ignorant he knows he is), the better his chances of being right?

There is plenty I don't know, and plenty I do know.  The best engineers I ever met were the ones that knew enough to tell the difference between "smoke and mirrors" and what might be true.  And were educated enough to determine the difference.

FYI, Einstien was sometime wrong, and he knew it.  He was just right enough and within the assumptions he made and stated.  Newton was sometimes wrong but plenty close enough for day to day stuff.  Knowing what can be neglected and still get useable results is important.

"Accepting that someone is right because he has a stack of publications behind his name, a few awards and perhaps a PhD, is very easy."  This can be dangerous, but a safer bet than a hipshot from an uneducated Aussie, even with an inquiring mind.  I should point out I have never met nor worked with a real Aussie, so I'm just speculating.

[Ray]

So, are you saying that the less one knows (more ignorant), and the less one knows he knows (the more ignorant he knows he is), the better his chances of being right?
[a href=\"index.php?act=findpost&pid=93545\"][{POST_SNAPBACK}][/a]

No. I'm saying that human pride and egotism can lead a person into believing he knows more than he actually does know and that the ability (or talent) to know that you don't know (rather than kid yourself that you do know, for all sorts of face-saving, status seeking, ego boosting reasons etc etc) is very much underrated.

[howiesmith]

No. I'm saying that human pride and egotism can lead a person into believing he knows more than he actually does know and that the ability (or talent) to know that you don't know (rather than kid yourself that you do know, for all sorts of face-saving, status seeking, ego boosting reasons etc etc) is very much underrated.
[a href=\"index.php?act=findpost&pid=93547\"][{POST_SNAPBACK}][/a]

So how does Ray know the difference between smoke and mirros and a humble but well published PhD ?

[Ray]

So how does Ray know the difference between smoke and mirros and a humble but well published PhD ?
[a href=\"index.php?act=findpost&pid=93548\"][{POST_SNAPBACK}][/a]

Ultimately, Howard, it's up to each of us to decide what makes sense, what's meaningful, what's useful, what's right or wrong. How we decide such matters is determined by our entire world experience, our education, our genetics, our environmnet and everything that has happened to us from the time, and including the time, that we were in the womb.

But to get back to what started this little diversion, both you and Bill seem to be implying that someone like Roger Clark has written a PhD thesis on the current state of CMOS developments in Canon's laboratories and that I, from a position of great relative ignorance, am disputing his conclusions.

Well, of course that's not true, is it. That's not what has happened. As I see it, Bill is using the credentials and comments of a qualified physicist who has a PhD gained some years ago and which is not necessarily directly relevant to cutting edge developments in CMOS imaging, in order to support what I consider to be a negative point of view with regard to future possibilities.

[bjanes]

However, as an Australian, I don't have much regard for authority   . There's a tradition here of throwing off the shackles of our imperial masters.
[a href=\"index.php?act=findpost&pid=93534\"][{POST_SNAPBACK}][/a]

It is us Americans that have fought two wars with the imperial masters, whereas (according to the US State Department notes) you Aussies still accept the British monarch as sovereign and still have the Union Jack on your national flag. In this historical context, we are more independent minded than you.  

Contrarians are occasionally correct, but if someone offered me an opportunity to invest in his perpetual motion machine, I would respectfully decline.

Bill

BTW, from what part of Australia do you come? Also, most of us have a high regard for the British, the above difficulties notwithstanding.

[Ray]

It is us Americans that have fought two wars with the imperial masters, whereas (according to the US State Department notes) you Aussies still accept the British monarch as sovereign and still have the Union Jack on your national flag. In this historical context, we are more independent minded than you. 

That's true, Bill. We didn't have to fight a war to gain our independence, but the psychological shackles are still there. It's interesting, perhaps comforting, that all 3 of us, the Americans, the British and the Australians, with their common heritage, are in Iraq. But very discomforting that this escapade seems a complete debacle. (Sorry! Completely off-track).

Contrarians are occasionally correct, but if someone offered me an opportunity to invest in his perpetual motion machine, I would respectfully decline.

Are you referring to the full-blooded aboriginal pictured on our $50 bill, David Unaipon, who had an obsession with perpetual motion? I see you've done your research well   .

BTW, from what part of Australia do you come? Also, most of us have a high regard for the British, the above difficulties notwithstanding.

I am British. I'm Australian as well, having emigrated here about 30 years ago.

[John Sheehy]

A car is traveling at one meter per second

An exposure is made of one second

The car has one meter of blur

Irrelevant of capture method / size of sensor / lenght of thread etc

With a higher resolution you will be able to appreciate the blur in greater detail

Tell me I am wrong ?   
[a href=\"index.php?act=findpost&pid=93502\"][{POST_SNAPBACK}][/a]

Simply put, with a higher resolution, your captured blur will more closely resemble the analog blur.  A lower resolution would confuse things further, either a tiny amount or a lot.