The "Megapixel Mess" - Part II

[Aku Ankka]

I believe it's time to bring back truth in advertising and publish optical resolution ratings rather than using the term “megapixel resolution” which has, defacto, multiple meanings depending on the individual type, size and configuration of sensor used in the equipment being considered.

The aveage consumer doesn't care about exact resolution figures or such. It is enough if there are competetive amount of megapixels there.

If you start publishing resolution figures, you'll be opening a new can of worms as the results are highly dependaple on the measuring methodology. Even if everything gets standardized, we'll have the problem of how to deal with the AA-filter as the proper workflow explicitely demands capture sharpening to reduce the blur. If we do not perform capture sharpening, we're not getting proper figures out of the test, but underperformance. The next question, is how much sharpening should be used - now the proper amount depends on the strength of the AA-filter. Should we just use typical unharp mask, or what if the manufacturer, knowing the point spread function of a optical assembly (lens, microlens, filters) to a reasonable accuracy, decides to effectively remove the negative effects of the AA-filter with deconvolution?

It is much simpler to all of us just to know the pixel count, and if one wants to know more details, one can always stydy the issue oneself - over 99.9% of the consumers don't want to.

I'm reminded of the confusion which followed Maxtor Corporations marketing decision to “reinvent” the definition of “megabyte.” Maxtor began marketing their hard disk drives as having “X” megabytes of capacity where megabyte was defined as one million pixels. The conveniently forgot the 48,576 additional bytes in a true

Well, maybe not pixels

Anyhow, it was the computer industry who decided first to redefine what mega-prefix (and kilo and giga) means, wasn't it

began defining megabyte the same way. Now that we have multi-gigabyte hard disk drives, this rather paltry 48,576 byte per megabyte omission becomes even more significant.

Surely you mean insignificant ; the relative difference does not change with larger hard drives, however, since most hard drives in the world are at most half full due to their sheer size, and because they're dirt cheap, it hardly makes any real difference if someone decides to re-redefine mega or giga back to it's original meaning

In my humble opinion, it's time for truth in advertising. Yes, only Sigma Corporation is really compromised by this issue right now. People who purchase high priced medium format cameras and Leica dSLR's which have no AA filter are not mislead and know what they are getting. The average consumer, however, is still confused and misled by what should be a very simple thing – tell it like it is and either report “accurate” LPH resolution figures or if there is a perceived need to use the terms megapixel and resolution as equivalent, then report the “actual” relationship. My D7000 Nikon then would be advertised as a camera with 10.2 megapixel resolution, 16 megapixel image file.

There are at most a few thousand medium format cameras sold each year with no AA-filter, maybe about that many Leicas as well, and maybe 10.000 Sigmas. There are 100's of millions of Bayer filtered cameras sold every year and those in general have either and AA-filter, or the pixel pitch is so small that diffraction acts as the source for anti-aliasing. Thus it would be rather silly to implement a new difficult to implement standard for reasolution measurement just to please a tiny crowd. In addition this new methodology would create pressure for the manufacturers to remove the AA-filters regardless of moire and other aliasing, or to cook the low-ISO raw-files with some deconvolution sharpening.

And btw, if you want honesty in advertising, you might want to tell Sigma to stop calling their cameras as 14 or 46 or whatever Mp cameras, and ask them to stop calling their cameras as being the only ones recording true colors and so on. Both are blatant lies, so far away from truth that no anti-alias filter could blur them away

Essentially, my 16 megapixel D7000 resolves about 20% less than “advertised” when described as having 16 megapixel “resolution.”

Essentially, a typical Foveon sensor camera resolves about 42% less than "advertised" when described as having 14 megapixel "resolution."

 

[Lin Evans]

that it is O.K. to misrepresent the capacity of a hard disk or the "resolution" of a digital camera as long as everyone does it? And that somehow it's less than O.K. for Sigma to misrepresents their optical resolution than for CFA manufacturers to misrepresent their optical resolution because Sigma misrepresents to a greater degree?

By the way - perhaps you should check your math and tell me how you arrived at 42% less "resolution" than advertised. Or, are you falling into the same trap by considering optical resolution equal to pixel count?? Compute the "true" optical resolution of a typical 14 megapixel CFA sensor or find yourself a good example of a resolution chart result of a typical 14 megapixel CFA sensor and show me how you arrived at your figure....

Hmmmm

Lin
--
learntomakeslidshows.net

 

[Paulo Ferreira]

Posted this on (part I) but got an edit snag
Here it goes again

They do even less then that. The lines you mention are black and white. If they were RGB coloured, the maximum resolution would be
((Total MPixel count):4) x sqrt(2)!!!
a 20MP CFA has a maximum full colour resolution of 7MP
The Hassi H4D-50 has a maximum ful colour resolution of 17.67MP
The Hassi H4D-50MS has a maximum full colour resolution of 50MP
The Hassi H4D-200MS has a maximum full colour resolution of 70MP

A 15MP Foveon (eg SD1) has ... 15MP full colour resolution. Pretty comparable and much cheaper.

There's a lot of "engineers" and corporations that found this MP count "standard" (no SI though not even Imperial) to be very effective for "kerching" $$$ purposes.

The science is lame though, the experiments are even lamer and the bias is astronomical. Still the "engineers" have to earn their salary and show they deserve it even if the self proclaimed experimental method is totally flawed.

--
--------------------------------------------------------

'Politicians and baby nappies should be changed regularly for exactly the same reason.'
Eça de Queirós (1845-1900)
-------------------------------------------
Paulo Ferreira
(equipment in profile)
http://www.pauloferreira.co.uk
http://pauloferreira.imagekind.com
http://www.pbase.com/pauloferreira
http://www.pbase.com/sigmadslr/paulo_ferreira

 

[Mostly Lurking]

Well, you're correct about writing code in Octal and Hexidecimal, but before that it was written in Machine Code. The first computer was built using relays and was 'programmed' by moving wires around. It was used to calculate artillery trajectories and occupied a whole house.

Prior to Intel's changing the definition, a CPU had to have 16-bit registers that could be paired to perform 32-bit operations, a 16-bit accumulator (Intel's had 2 8-bit accumulators that could be paired for operations), and I forget what else. The issue isn't of sufficient importance for me to provide references---and neither is your (plural) mistaken beliefs on the subject. By the way, Motorola's MC68000 was a a true 16-bit processor, not the 32-bit device it's now claimed to be.
--
William Wilgus

 

[Daniel Browning]

I believe it's time to bring back truth in advertising

Ha ha ha! I Thanks for the laugh -- I need that on Mondays. There is no truth to bring back, because truth has never been a part of advertising. From today's camera manufacturers to PT Barnum's newspaper ads, perhaps going back to the first cave painting (probably timeshare ads). Truth is the antonym of advertising.

Do you know Satan's full job title? "Lucifer, Father of Lies, Director of Marketing." Even if a marketer/advertiser somehow knew, beyond a shadow of a doubt, that truth-in-advertising would quadruple profits, he would still lie . He has to -- it is his nature. Like a wolf spider consuming its young, the ad man spinning a web of deceit is innate, instinctual behavior.

and publish optical resolution ratings

I think the single most useful resolution measurement is a zone plate. It shows resolution as well as aliasing, and can be easily converted to an MTF chart and easily digestible numbers (e.g. MTF-50). But since manufacturers can't even be bothered to publish honest MTF charts (e.g. Canon's physically-impossible ones), there's little hope of getting good zone plates.

--
Daniel

 

[Joseph S Wisniewski]

Re: To my dis-believer(s):

It's one word, "disbelievers". And it's definitely plural, because you have no believers, at all.

[paragraph that a 12 year old could have copied from any one of thousands of articles on the history of computing, deleted].

Prior to Intel's changing the definition, a CPU had to have 16-bit registers that could be paired to perform 32-bit operations, a 16-bit accumulator (Intel's had 2 8-bit accumulators that could be paired for operations), and I forget what else.

That is all totally incorrect.

The issue isn't of sufficient importance for me to provide references

The issue is that you have zero credibility, and that everyone here considers you to be a troll. Now, if you believe that issue is not "of sufficient importance" for you to address, it specks volumes on both your own lack of self respect and social skills and your opinion of the other (and obviously better educated) members of this forum.

Even your name is a lie: "Mostly Lurking". You're averaging about 24 posts a week, and just hit your thousandth post in 9 months on this relatively small Sigma forum,which makes you one of the most prolific posters in the place. If that's "lurking", I'd hate to see you when you "participate".

I guess a better name for you would be "Frequently Trolling".

and neither is your (plural) mistaken beliefs on the subject.

No one has been shown to have made any mistakes on the subject other than you.

By the way, Motorola's MC68000 was a a true 16-bit processor, not the 32-bit device it's now claimed to be.

By the way, no one is interested in any more of your inane claims.

Provide some references to your prior claims, or go away.

Oh, I'm sorry. I used some words that were 2 or 3 syllables, and a compound sentence. That may be beyond your ability to deal with. Here, I'll rephrased it in monosyllables.

Put up or shut up.

--
Rahon Klavanian 1912-2008.

Armenian genocide survivor, amazing cook, scrabble master, and loving grandmother. You will be missed.

Ciao! Joseph

http://www.swissarmyfork.com

 

[Aku Ankka]

that it is O.K. to misrepresent the capacity of a hard disk or the "resolution" of a digital camera as long as everyone does it? And that somehow it's less than O.K. for Sigma to misrepresents their optical resolution than for CFA manufacturers to misrepresent their optical resolution because Sigma misrepresents to a greater degree?

There is no misrepresentation of any importance when it comes to making megapixell statemements unless one lies about the numer of them. Using megapixels is a relatively good way of telling about one property of the camera, especially since almost all the cameras out there use directly comparable sensors. It's not perfect, but it is enough for 99.9% of the consumers. Those who care more, as you obviously do, can figure out more details themselves.

And Sigma's cameras are still absolutely not relevant for megapixel parketing issue. They sell maybe 10.000 a year. Hundreds of millions of Bayer CFA equipped cameras are sold a year. Why bother to make a complicated and potentially unsolvable system for resolution representation when the current one works fine for 99.99% of cameras out there?

What alternative would you propose? How would you avoid the pitfalls in that alternative? Do you think it would be simple and easy enough for my now deceased grand mother to understand?

By the way - perhaps you should check your math and tell me how you arrived at 42% less "resolution" than advertised. Or, are you falling into the same trap by considering optical resolution equal to pixel count?? Compute the "true" optical resolution of a typical 14 megapixel CFA sensor or find yourself a good example of a resolution chart result of a typical 14 megapixel CFA sensor and show me how you arrived at your figure....

1-sqrt(4.7) / sqrt(14)

There is no trap when talking about a single sensor/camera. Sigma advertises it's cameras as having three times as many pixels as they actually have, ie. they lie about the resolution that miuch. What the sensors of the rest of the world do is absolutely irrelevant for this.

Anyhow, just for the record, the Foveon sensor without the AA-filter is only slightly better at resolving than a Bayer sensor without the AA-filter and using modern demosaicing algorithms. The difference is only about 10%, give or take a few percentage depending on the image (colors and amount of light). What often confuses people is that the comparisons are almost always made against AA-filtered cameras and the beyong Nyquist false detail is considered detail which it of course is not (although it may appear rather pretty in practise).

Anyhow, the AA-filter can make quite a difference, especially on older SLRs with strong ones.

 

[Lin Evans]

Hi Daniel,

Thanks for the comments. Unfortunately, you are not old enough to remember when there "was" generally truth in advertising. I am.

I've known plenty of V.P. and Director's of Marketing who went out of their ways to further truth in advertising. Sorry that you are so jaded by those who have "spun the tale." Yes, there have always been "hucksters," but there have been plenty of truthful folks selling good products and being entirely honest with their customers.

Best regards,

Lin

I believe it's time to bring back truth in advertising

Ha ha ha! I Thanks for the laugh -- I need that on Mondays. There is no truth to bring back, because truth has never been a part of advertising. From today's camera manufacturers to PT Barnum's newspaper ads, perhaps going back to the first cave painting (probably timeshare ads). Truth is the antonym of advertising.

Do you know Satan's full job title? "Lucifer, Father of Lies, Director of Marketing." Even if a marketer/advertiser somehow knew, beyond a shadow of a doubt, that truth-in-advertising would quadruple profits, he would still lie . He has to -- it is his nature. Like a wolf spider consuming its young, the ad man spinning a web of deceit is innate, instinctual behavior.

and publish optical resolution ratings

I think the single most useful resolution measurement is a zone plate. It shows resolution as well as aliasing, and can be easily converted to an MTF chart and easily digestible numbers (e.g. MTF-50). But since manufacturers can't even be bothered to publish honest MTF charts (e.g. Canon's physically-impossible ones), there's little hope of getting good zone plates.

--
Daniel

--
learntomakeslidshows.net

 

[Lin Evans]

that it is O.K. to misrepresent the capacity of a hard disk or the "resolution" of a digital camera as long as everyone does it? And that somehow it's less than O.K. for Sigma to misrepresents their optical resolution than for CFA manufacturers to misrepresent their optical resolution because Sigma misrepresents to a greater degree?

There is no misrepresentation of any importance when it comes to making megapixell statemements unless one lies about the numer of them. Using megapixels is a relatively good way of telling about one property of the camera, especially since almost all the cameras out there use directly comparable sensors. It's not perfect, but it is enough for 99.9% of the consumers. Those who care more, as you obviously do, can figure out more details themselves.

And Sigma's cameras are still absolutely not relevant for megapixel parketing issue. They sell maybe 10.000 a year. Hundreds of millions of Bayer CFA equipped cameras are sold a year. Why bother to make a complicated and potentially unsolvable system for resolution representation when the current one works fine for 99.99% of cameras out there?

What alternative would you propose? How would you avoid the pitfalls in that alternative? Do you think it would be simple and easy enough for my now deceased grand mother to understand?

Simply advertise optical resolution in LPH for both color and black and white. This would be a "truthful" representation of "resolution" and far closer to reality than equating pixel count to sensor resolution.

By the way - perhaps you should check your math and tell me how you arrived at 42% less "resolution" than advertised. Or, are you falling into the same trap by considering optical resolution equal to pixel count?? Compute the "true" optical resolution of a typical 14 megapixel CFA sensor or find yourself a good example of a resolution chart result of a typical 14 megapixel CFA sensor and show me how you arrived at your figure....

1-sqrt(4.7) / sqrt(14)

That's what I thought. This is an invalid figure. To accurately measure the difference as a percentage you would need to divide LPH of Sigma's 4.7 mp sensor by by LPH of the 14 mp CFA sensor it is being compared with.

There is no trap when talking about a single sensor/camera. Sigma advertises it's cameras as having three times as many pixels as they actually have, ie. they lie about the resolution that miuch. What the sensors of the rest of the world do is absolutely irrelevant for this.

Anyhow, just for the record, the Foveon sensor without the AA-filter is only slightly better at resolving than a Bayer sensor without the AA-filter and using modern demosaicing algorithms. The difference is only about 10%, give or take a few percentage depending on the image (colors and amount of light). What often confuses people is that the comparisons are almost always made against AA-filtered cameras and the beyong Nyquist false detail is considered detail which it of course is not (although it may appear rather pretty in practise).

Removing the AA filter has no bearing on the color resolution differences which are variable depending on color for CFA sensors. It's not possible to give a single figure, but would be necessary to measure using a color resolution chart. The Foveon sensor has full resolution regardless of color; the CFA sensor does not.

My very good resolving Nikon D7000 has essentially 20% less optical resolution than one would assume when the terms "megapixel and resolution" are used as equivalents in advertising. For example, "a 16 megapixel resolution" D7000 produces about 26 LPH measured b&w resolution from a 3264 vertical pixel sensor. To be truthful, one would have to say that it is a 10.2 megapixel resolution camera. Having carefully measured my Sigma SD15 via resolution charts, I find that it measures very close to 1760 LPH which is the theoretical limit.

When I run the numbers for the 4.7 mp sensor versus a very good Bayer/CFA camera with a 14 megapixel sensor, I end up with 1760/2400 or 26.67% less b&w resolution than the "14.2 megapixel" Nikon D3100. That's quite a ways from your 42% calculation.

Anyhow, the AA-filter can make quite a difference, especially on older SLRs with strong ones.

Yes, there are considerable differences depending on individual cameras. My first six megapixel dSLR was a Kodak DCS-460 with a removable AA filter. The resolution differences were quite visible on prints with and without the AA filter.

--
learntomakeslidshows.net

 

[Makinations]

Simply advertise optical resolution in LPH for both color and black and white. This would be a "truthful" representation of "resolution" and far closer to reality than equating pixel count to sensor resolution.

That's a horrible measure. What if the end user doesn't buy the optics that goes with that measure? You going to have LPH listed for how the product performs with each kit lens?

Pixel count is as good as anything. There is only one freak sensor it doesn't work for.

 

[Erik Magnuson]

Prior to Intel's changing the definition,

When and how did Intel change a definition that predates Intel microprocessors? For example, the PDP-8 was a "12-bit" architecture. The PDP-11 was a 16-bit architecture.

a CPU had to have 16-bit registers that could be paired to perform 32-bit operations, a 16-bit accumulator

Please cite a source for this and then explain why Data General and DEC were doing all wrong for 10 years before the 68000.

The issue isn't of sufficient importance for me to provide references

You don't want to let the facts get in the way of what you "know", so you can't be bothered to check your references. Hmm.

--
Erik

 

[Lin Evans]

Horse pucky!

Lin

Simply advertise optical resolution in LPH for both color and black and white. This would be a "truthful" representation of "resolution" and far closer to reality than equating pixel count to sensor resolution.

That's a horrible measure. What if the end user doesn't buy the optics that goes with that measure? You going to have LPH listed for how the product performs with each kit lens?

Pixel count is as good as anything. There is only one freak sensor it doesn't work for.

--
learntomakeslidshows.net

 

[Mostly Lurking]

If you had any brains, you'd have realized that and why. Bye-bye, unjustifiably arrogant!

--
William Wilgus

 

[Mostly Lurking]

--
William Wilgus

 

[Erik Magnuson]

and it appears I should be saving my breath.

But here is a well respected source:
http://jbayko.sasktelwebsite.net/cpu.html

I remember the pre-WWW days when this was distributed via usenet. Man, am I getting old.

--
Erik

 

[Joseph S Wisniewski]

Th ere was a reason the 's' was in parens.
If you had any brains, you'd have realized that and why.

Hmmm, maybe it was the same reason you:

• Put a space in "There".
• Put a hyphen in "disbeliever".
• Unnecessarily capitalized "Octal", "Hexadecimal", and "Machine Code".
• Misspelled "hexadecimal".

http://en.wikipedia.org/wiki/Functional_illiteracy

Bye-bye, unjustifiably arrogant!

And goodbye to you, too.

I take it you're saying goodbye because you're leaving dpReview? Because I'm not leaving.

But once again, you are getting the roles confused. It is you who have been consistently in error, you who are "unjustifiably arrogant". I'm sorry that pointing the truth out to you disturbs you so.

There are professionals that can help with your problem. You should seek them out.

Goodbye again.

--
Rahon Klavanian 1912-2008.

Armenian genocide survivor, amazing cook, scrabble master, and loving grandmother. You will be missed.

Ciao! Joseph

http://www.swissarmyfork.com

 

[Makinations]

So why is the lens bit wrong?

Lin

Simply advertise optical resolution in LPH for both color and black and white. This would be a "truthful" representation of "resolution" and far closer to reality than equating pixel count to sensor resolution.

That's a horrible measure. What if the end user doesn't buy the optics that goes with that measure? You going to have LPH listed for how the product performs with each kit lens?

Pixel count is as good as anything. There is only one freak sensor it doesn't work for.

--
learntomakeslidshows.net

 

[Ed_S]

Lin,

Thanks for posting so clearly what I was "poking at" in another recent thread. I've lost track of which thread, but in it a newcomer (IIRC) was reopening the issue of the Foveon 3X megapixels vs. CFA. My short and unsophisticated attempt to defuse the argument was simply to point out (sans mathematics) that a CFA of X MP generally does not resolve at that level (as you've eloquently pointed out) and that the Foveon way of counting just comes closer to reality than the megapixel myth.

Bottle it and spread it around!

Best,
--
Ed_S

http://www.pbase.com/ecsquires