Pixels Are Like Cupcakes. Let Me Explain.

By Russ Juskalian, The New York Times

It happens to all of us: the moment when one finds out that more megapixels and better photographs aren’t always the same thing. To be disabused of the Megapixel Myth—this decade’s analog of the Megahertz Myth—can lead to an existential buyer’s crisis in miniature.

Disbelief, at first, gives way to a sort of embarrassing self-questioning: You mean, 15 megapixels isn’t three times better than 5 megapixels? This year’s model isn’t better than last year’s? I spent all that money upgrading—for nothing?

READ ON at nytimes.com

________________________

Mike  (Thanks to Aaron Dill)

Featured Comment by Robin P: "Well, it read like an advert for dpreview to me but I sent the link to a technophobe friend anyway and he replied 'Very interesting and easy for a novice like me to understand.'

"All grist to the mill in the struggle to make technology easier for the man in the street to understand and deter the manufacturers from persuading us to buy what we don't really need."

Featured Comment by Mike Jones: "The article brings one additional factor into the equation, but is little more enlightening than the pixel counting scenario. We are still judging digital photography within the digital realm, simply because it is digital. Pixels, sensor size, pixel density, ISO etc. When did one ever judge a photograph by such criteria? If so where would 35mm been in comparison to large format? Colour rendering, sharpness, gamut, contrast, bokeh are all factors too, but not so easily quantified. Especially when related to the final image."

Comments

[Jason]

Thanks, now I am craving a cupcake! (And I've already reached my morning sugar limit from the coffee...)

Good article though, and it well illustrates the subtle suggestion that it's not the gear, it's what's behind the gear (the photographer) that matters...

[Dennis Allshouse]

This, of course, is the original reason for moving from APS to 35mm sized sensors: get more Mpixels without making individual pixels too small. Similarly for going on to "medium" format sensors.

Now in both 35mm and medium format the pixel wars have pushed pixel sizes close the formerly 'bad' zone. I'm being intentionally vague with the numbers here. Interestingly the Times article didn't attempt to address the how small is too small question. Does anyone have an answer for this? Or is it a moving target ie. today 5 microns too small, tomorrow after some hypothetical advance 4 micron is ok? Clearly for some too much is never enough. (tongue in cheek: after all, what serious, demanding, only the highest quality will do, photographer prints anything smaller than 24x36 inch prints!)

[Christopher Lane]

I found this article to be a good explanation of the subject. As you know, I am a big fan of the Nikon D700 and now I think I know why. According to DP Review's numbers, it has the lowest pixel density of any DSLR.

1.4 Nikon D3/D700
2.1 Leica M8
2.4 Canon 1 Ds Mark III, 5D Mark II
2.9 Sony a900

Chris

[charlie]

Good gear may not make the photographer, though some people and circumstances demand it. You can't really argue that much about it.

I know one thing for sure... in just about every way, the Nikon D700 trounces my Olympus E300 that I still happily use every day. Almost every day.

I for one am glad I never made a miniscule "upgrade" to a better camera in the past couple years. I know that when I do it will be a considerably more enjoyable camera to use. I also know it will be FF regardless of how many pixels.

Nice pedestrian written article that I could understand.

[JAlan]

Hi Mike,
We who routinely visit your site, as you know, are pretty passionate about photography. Furthermore, many of us are technically (myself included) into cameras.....no....facinated about cameras, and be that as it may, I can't help but say after reading this article and the former one concerning the epic Red cameras below, and also having just seen Michael's review of the future of the Red system and what it MAY be capable of in the future, I sit back and can't help but think that as technology proceeds at it's phenomenol rate and how cameras have relatively speaking, have developed at an exponential rate, I can't say the phtographic results have both artistically and creatively have grown exponentially. Meaning, technology has far exceeded th ecapabilities of 99.9% of photographers. Don't get me wrong. I believe brilliant work is being created today but most of the work by the masses using this phenomenal equipment is pretty mediocre......

On another note, concerning technology, check this out.....it's pretty sobering...

http://www.youtube.com/watch?v=jpEnFwiqdx8

Best Regards,
Alan

[Ctein]

Dear Dennis,

One size does not fit all, and, yes, it's a moving target.

Just as there will be photographers who insist that 30 MP is the bare minimum they could tolerate, others will demand 15 stops of exposure range or noise-free images at ISO 50,000.

But for REASONABLE demands, currently 5 microns is not too small. Actually, my Fuji S100, with 2.5 micron pixels, has only one stop worse performance than a Nikon D200. If the same technology was available with 3.5 microns pixels (it's not) it'd be just as good as the D200 was.

Sensor design also continues to improve. Doesn't even need any hypotheticals to get a lot better, just implementing stuff that's currently in the lab/development phase.

Pixel density matters, but it's not half as important as people are making it out to be. Its just the new kid on the block. Like when stereo buyers suddenly figured out that it wasn't about raw wattage; distortion mattered. And then they all became distortion-mad for a while, to the exclusion of everything else.

You can't buy a stereo on specs; neither can you choose a camera that way.

pax / Ctein

[mu23]

Most of the dozen or so reviews I've read so far, say the 40D and 50D are equal in image quality and noise characteristics. Owning both models, I tend to agree with the majority. The Times has been grinding this axe a little too hard and a little too often for my liking. Should we have stopped dSLR resolution at 5mp? I think not.

[BAG]

I believe that the push for higher and higher pixel counts comes more from the marketing department of the camera companies than anywhere else. I might be wrong but this most likely evolved in the early days of digital cameras when a jump from 1Mp to 2Mp for instance was a huge increase in potential print size and picture quality and therefore encouraged higher sales.
When this was translated to consumer point and shoots a gullible public was convinced that higher pixel count meant better results, even though 99% of them either don't make prints or only print postcard size.

[Noons]

BAG: you are absoltely right.

The ONLY reason for the push into higher and higher MPs is uniquely because technically ignorant marketeers have identified that as an easy "more is better" approach.

There is absolutely no need whatsoever for consumer grade - or indeed even "prosumer" - cameras, be they dp&s or dslr, to have more than 12MP. Anything over that is clearly an overkill and achieves nothing else than increase noise levels.

Light does not increase or decrease its wavelength to match the needs of marketing.

Sensels at 5microns are already reaching the maximum ability of light itself to stimulate a response in an electronice device of the type used to capture photon impacts.

The solution is the same film found decades ago: get a bigger sensor (larger frame).
Not more sensels!

[Kobi Eshun]

One of my favorite distractions is playing The MegaPixel Challenge (tm) with unsuspecting co-workers:

"Hey, guess how many megapixles for this picture," starting out with a random Canon D30 shot. "How about this," following up with something from a Panasonic LX2. As you might imagine, most have trouble believing the D30 output is "only" 3 megapixels, compared to the LX2's 10.

Some get it when I point out that my 20-inch widescreen LCD at 1680x1024 only displays 1.6 megapixels. But that doesn't stop them from muttering disdainfully about my old faithful DSLR. Oh well.
--
kobi

[Craig Norris]

The author doesn't quite get it right where he states:

"The more photons a cup catches, the brighter that cup’s color. Totally empty cups record black; totally full cups record white."

Totally empty cups will record an absence of the corresponding primary colour. Totally full cups will record a maximum brightness for the corresponding primary colour. The resulting colour of the pixels in the photograph as perceived by the viewer depends on how empty or full the neighbouring cups also happen to be.

Only when the neighbouring cups are also full or empty can one say the pixels will be "white" or "black", since it requires three primary colours, and therefore at least three 'cups' to determine a colour, even if the colour is white or black.

I don't mean to seem anally pedantic - I just hope to prevent a misunderstanding by those readers who aren't familiar with the technology, and there's no doubt that the article is written for those people.

[Martin Doonan]

I think camera salesman should be forced to read & understand this stuff, too. The amount of guff I hear from them regarding megapixels when selling basic cameras is astonishing. The biggest selling point being used for more megapixels is bigger prints (becasue, of course, print is only a matter of resolution!).

[John Roberts]

I learned by experience that you can't always trust the "common wisdom" of the reviewers. When I bought my first digital camera, a 4MP Canon G3, the "experts" said it was good for producing prints up to 8x10, but no larger. As an experiment, I had an 11x14 of one of my G3 photos made at MPIX, just to see how my little G3 would do at that size. When I got it back, I couldn't believe how great it looked. In all respects, it was superior to any 11x14 print I had ever had made from a 35mm slide or negative. So I decided to try a slightly up-rezzed 16x20. Same result: beautiful at a normal viewing distance! I eventually upgraded to a dslr for the features, but MP counts haven't impressed me since I saw for myself what could be done with a "little" 4MP image.

[Ctein]

Dear Craig,

You have it backwards. Pixels only record luminance levels. It's technically correct to describe their response range as being from "black to white".

Color is synthesized later, based on the filter and response patterns.

There are photon-counting sensors that can record color (wavelength) directly. The ones in your camera, though, only record brightness.

If I were limited to 25 words or less (remember, it's a newspaper column!) I'd write much what the author did, for lay audience. It ain't technically perfect, but it's close enough and it beats what's in second place.

Pendantry is usually not worth the energy you invest on it.

pax / Ctein

[Ctein]

Dear Noons,

Wholehearted agreement that MP Mania is just another version of the horsepower (or peak watts) race.

I have some very impressive 8x10's made with a Kodak 2 MP camera from last century; folks would be amazed what well-applied pixels can do.

That said, we are not anywhere close to physical noise limits with 5 micron pixels, and there are substantial gains in overall image quality that can be achieved with smaller pixels. Sensor and filter array design are nowhere close to optimized, yet. And there are some VERY useful design tricks one can exploit if one gets down near the 1 micron range in a decent-sized sensor.

(And, no, I'm not assuming radically different sensor technologies.)

Arguing right now that some pixel size of X is the fundamental limit for quality would have been like arguing back in the 1970's that the film speed/grain performance ratio had reached its limit. We are nowhere near close enough to the endpoint to slap a number on it.

pax / Ctein

[Craig Norris]

Hello Ctein.

I beg to differ about what a pixel (sensor photosite) records.

"Pixels only record luminance levels. It's technically correct to describe their response range as being from "black to white"."

Pixels on the sensor, the so-called 'cupcakes', record in the main only the photons that made it through a colour filter. Therefore each pixel's measured accumulated charge only indicates the brightness or luminosity of light that corresponds to the wavelength of a particular primary colour - red, green or blue. We therefore shouldn't call the measured outcome for each pixel (photosite) on the sensor as black or white. The direct outcome at the A to D converter can only be a measurement of the amount of red or green or blue.

Our perception of colour in a photograph is based on the final RGB mix, not on the value of one primary colour only. Each pixel on the sensor (a photosite) only detects one primary colour. It therefore can never be called 'white'.

We can't call any pixel black or white until after the demosaicing process, by which time the word 'pixel' has a different meaning.

There still exists in our industry some confusion about what a "pixel" really is. But in the article, the author makes it clear with his 'cupcake' metaphor that he means an individual photosite on the sensor.

Your statement would be true, Ctein, if it was about the coveted true monochrome digital camera - a digital camera without an RGB mosaic filter.

May Santa Claus bring us that monochrome digital camera soon.

[Mike J.]

Craig,
In a classic tricolor camera, there was a beam splitter and filters that allowed three ortho B&W plates to be exposed from one scene.

http://www.loc.gov/exhibits/empire/making.html

Would you consider the plates to be recording color or only luminance?

Mike J.

[Noons]

This is not the best avenue for a resolution debate, so I'll refrain from further replies after this one.

I think I know what you mean, Ctein. Although I tend to stay away from "hidden tricks no one is privvy to except the chosen few", when talking about sensors.

Fact is: the very best consumer lenses can resolve maybe 100line pairs per millimetre, on a good day, with a good tripod. Of course there is better, I'm talking about consumer level equipment!

1 micron sensels - yeah, I know you need a lot more than 1 sensel for a true image pixel, that's not the point - is waaaay beyond what a consumer lens can resolve with any quality. Even 3 microns is borderline.

You state that improving to 1 micron can be done without major changes to current technology, while providing the example of film in the 70s versus modern film as a parallel.

Well, not quite: film from the 70s is technologically completely different from modern t-grain or other more advanced formulas. Which are useless unless accompanied by improvements across the board.

And that is also what needs to happen to sensors and cameras as a whole for smaller sensels to be relevant or even useful.

My point is that simply reducing the size of the sensel to increase resolution achieves preciously nothing by itself.

It needs to be accompanied by corresponding increases in lens resolution, as well as increased noise processing mechanisms and algorithms if we are to shoot at anything over 100ISO.

That is not happenning in the mad rush for the single target of "more pixels". Want proof? Pick ANY of the 12MP and over consumer p&s cameras and none of them has any image quality left after 400ISO. While other models with larger sensels can easily handle 800, even with outdated image processing.

Sure: one can downress a 12MP image to 6MP and make it look good. What exactly has one gained then from taking the image at 6MP in the first place? Nothing.

And that is the problem.

[Ctein]

Dear Craig,

Wavelength is not color. Color is a construct, both in the camera and in the human eye. You can talk about the pixel recording the brightness of a specific wavelength band; you cannot meaningfully talk about it recording color. It doesn't.

You're hung up on the phrase "black to white." It's a minor imprecision. A much bigger one is giving people the idea that a single pixel on a photosensor records "color."

Pixel is a well-defined term for most players in the industry, except for the Foveon people who tried to redefine it to their advantage. It's a term of geometry-- it refers to the smallest region of information. Pixels can have a wide variety of bit depths, and what data's associated with a pixel depends upon the physical device and the application.

pax / Ctein

[bobdales]

To my old, addled mind it boils down to this:
1. Other things being equal a bigger sensor allows a bigger print.
2. Other things are never equal.

Eschew pedantic obfuscation and pity the poor marketing guy who has to hang his message on a single metric. That's why there will always be things like horsepower, megaherz, gigabytes, mini-this, maxi-that, and of course megapixels.

bd

[Craig Norris]

Mike,

In a "3-plate" or "3-tube" or "3-chip" camera, the sensors are recording the luminosity of a primary colour, just like the 'cupcakes' of a Bayer-pattern sensor. That's different to "luminance" in it's usual meaning.

The three primary colour luminosities are later mixed in order to arrive at a final colour image, which can then be further analysed as separate but companion 'luminance' and 'chrominance' signals.

Craig

[Craig Norris]

Ctein wrote: "Color is a construct, both in the camera and in the human eye. You can talk about the pixel recording the brightness of a specific wavelength band; you cannot meaningfully talk about it recording color."

Exactly, Ctein. That's my point.

The sensor pixel simply generates a measurable charge that can be used to indicate the brightness of the primary colour (red, green or blue) that falls on that pixel.

Craig

[Ctein]

Dear Noons,

"the very best consumer lenses can resolve maybe 100line pairs per millimetre, on a good day, with a good tripod."

Ummm, not according to any test data I've seen (or collected). The best lenses peak out at 300 lp/mm, give or take. A lens which tops out at 100 lp/mm is actually pretty mediocre by today's standards.

You read my 1 micron remark backwards. I'm not saying the tricks get you to 1 micron, I'm saying there are very useful tricks and methods you can apply to image capture with 1 micron pixels that you can't do with large ones.

The comment about film was meant to provide perspective, not technological parallel. Let me say it without, then-- if you really think you know the lower limit to useful pixel size, you're just guessing and you're likely wrong.

Now, there's nothing wrong with making guesses in the matter; we all do. But you're presenting it as if it were established fact.

pax / Ctein

[Kevin McLoughlin]

I remember being instantly struck by the superb image quality of the original Canon D30 (3.1 mpx) when it was released in 2000.
I think Canon struck just about the right balance of pixel size in the D30.
A full frame equivalent cmos sensor at approx 8 mpx would be irresistible for me.