Shopping For Pixels
by Ctein
In the spirit of the season (which is all about cool toys), here's my consumer advice for the spec-obsessed camera buyer:
Too many buyers are way too interested in pixel counts. I'd hoped consumers would have learned better, but the latest feeding frenzy over 10 megapixel (vs. 8 megapixel) cameras sadly proves otherwise.
Pixel counts matter less than some think. Image resolution goes as the square root of the number of pixels. A 10-megapixel photograph has only 12% more resolution than one of 8 megapixels, all other things being equal (they're usually not). It's almost impossible to see a 12% difference under laboratory conditions with carefully designed test scenes. Realistically, you might just discern the difference between an 8-megapixel and a 10-megapixel image, but probably not.
A 10-megapixel photograph has only 12% more resolution than one of 8 megapixels, all other things being equal
We're easily fooled by numbers. People will pay much more for a 250-watt home sound system than a 200 watt one, because they think they're getting a lot more sound. Any audiophile will tell you that's only a couple of dB. We're seduced by CPU clock speeds; look at the huge premium manufacturers command for a new processor that runs at, say, 2.8 gigahertz instead of 2.4 gigahertz. In otherwise-matched systems, that will gain you less than 10% on throughput. Is it really that important to be able to run your Photoshop action in 55 seconds instead of 60 seconds?
Worse, things are never equal. A 250-watt sound system with high peak distortion isn't half so good as a 200 watt one with low peak distortion. A 2.8 gigahertz processor on a slow mother board with slow RAM and an IDE hard drive will poke along compared to the slower processor supported by a fast board and RAM and a SATA drive.
Different digital cameras extract very different amounts of sharpness and fine detail from their pixels. There are 8- (even 6-) megapixel cameras that produce sharper photographs than the poorest 10-megapixel ones. Raw pixels don't count. It's what you do with them.
Dig deeper into the specifications. Find comparison tests that tell you real camera resolution in lines. Learn whether those extra pixels are actually getting you extra detail or you're just paying for a big number. And, remember to take a close look at noise levels, color accuracy, and exposure capture range. They're worth more to you than a smidge more sharpness.
Ctein responds: That's an excellent question, and there's a lot of complicated information that goes into the answer. Whole articles have been written. Forgive me for just presenting results here without providing any supporting data.
First, sending more pixels to the printer doesn't produce worse results than sending fewer. Don't worry about that. You don't need to downsize anything before printing.
As for how many pixels are enough, consider a 10 megapixel camera image being printed out as a 7 in. by 10 in. photo. An average 10-megapixel camera actually has a resolution roughly equivalent to 3.5 megapixels. In a print that will work out to (insert mathematical arm-waving here) about 4 lp/mm.
That is a nice, decently-sharp print! But it's not the limit of what we can see; good eyes under good viewing conditions can resolve detail twice that fine or better. If you put a print that resolves 8 lp/mm next to one that results 4 lp/mm, you will see the improvement in sharpness. Kind of the same way putting a really sharp 35mm film print next to a really sharp 8x10" film print shows the difference. There's nothing wrong the 35mm print! It's just the 8x10 format print looks even sharper.
It would take a 40-megapixel camera to get you to that 8 lp/mm in a print (and that's not even the limit of what we can see). We have a long way to go before running out of sharpness improvements.
Do you need a 40-megapixel camera? No more than you need an 8x10 view camera. Would it be visibly better? Yes. Would it hurt? No.
In the spirit of the season (which is all about cool toys), here's my consumer advice for the spec-obsessed camera buyer:
Too many buyers are way too interested in pixel counts. I'd hoped consumers would have learned better, but the latest feeding frenzy over 10 megapixel (vs. 8 megapixel) cameras sadly proves otherwise.
Pixel counts matter less than some think. Image resolution goes as the square root of the number of pixels. A 10-megapixel photograph has only 12% more resolution than one of 8 megapixels, all other things being equal (they're usually not). It's almost impossible to see a 12% difference under laboratory conditions with carefully designed test scenes. Realistically, you might just discern the difference between an 8-megapixel and a 10-megapixel image, but probably not.
A 10-megapixel photograph has only 12% more resolution than one of 8 megapixels, all other things being equalWe're easily fooled by numbers. People will pay much more for a 250-watt home sound system than a 200 watt one, because they think they're getting a lot more sound. Any audiophile will tell you that's only a couple of dB. We're seduced by CPU clock speeds; look at the huge premium manufacturers command for a new processor that runs at, say, 2.8 gigahertz instead of 2.4 gigahertz. In otherwise-matched systems, that will gain you less than 10% on throughput. Is it really that important to be able to run your Photoshop action in 55 seconds instead of 60 seconds?
Worse, things are never equal. A 250-watt sound system with high peak distortion isn't half so good as a 200 watt one with low peak distortion. A 2.8 gigahertz processor on a slow mother board with slow RAM and an IDE hard drive will poke along compared to the slower processor supported by a fast board and RAM and a SATA drive.
Different digital cameras extract very different amounts of sharpness and fine detail from their pixels. There are 8- (even 6-) megapixel cameras that produce sharper photographs than the poorest 10-megapixel ones. Raw pixels don't count. It's what you do with them.
Dig deeper into the specifications. Find comparison tests that tell you real camera resolution in lines. Learn whether those extra pixels are actually getting you extra detail or you're just paying for a big number. And, remember to take a close look at noise levels, color accuracy, and exposure capture range. They're worth more to you than a smidge more sharpness.
Posted by: CTEIN
Question from dingbat:
I'm still confused by conflicting advice on this issue. Doesn't it all
come down to print size? Yes, a 10-MP camera nominally has 12% more
resolution than 8 MP, but isn't that meaningless on an 8x10 print? In
fact, wouldn't a 36-MP image look the same as a 6- MP image on an 8x10
print—maybe even worse due to the downsizing?Ctein responds: That's an excellent question, and there's a lot of complicated information that goes into the answer. Whole articles have been written. Forgive me for just presenting results here without providing any supporting data.
First, sending more pixels to the printer doesn't produce worse results than sending fewer. Don't worry about that. You don't need to downsize anything before printing.
As for how many pixels are enough, consider a 10 megapixel camera image being printed out as a 7 in. by 10 in. photo. An average 10-megapixel camera actually has a resolution roughly equivalent to 3.5 megapixels. In a print that will work out to (insert mathematical arm-waving here) about 4 lp/mm.
That is a nice, decently-sharp print! But it's not the limit of what we can see; good eyes under good viewing conditions can resolve detail twice that fine or better. If you put a print that resolves 8 lp/mm next to one that results 4 lp/mm, you will see the improvement in sharpness. Kind of the same way putting a really sharp 35mm film print next to a really sharp 8x10" film print shows the difference. There's nothing wrong the 35mm print! It's just the 8x10 format print looks even sharper.
It would take a 40-megapixel camera to get you to that 8 lp/mm in a print (and that's not even the limit of what we can see). We have a long way to go before running out of sharpness improvements.
Do you need a 40-megapixel camera? No more than you need an 8x10 view camera. Would it be visibly better? Yes. Would it hurt? No.
22 Comments:
In a past life, a PC maker told me that *corporate* buyers would pick a faster clock-rate laptop over a slower-cpu-clock-rate laptop. And that they made this choice after said OEM *told them* the slower-cpu-clock-rate laptop would be a faster machine.
There must be some very deep human fascination with The One Number Of Merit.
Yep, and even when the more pixels are actually better than the fewer, U still need to be running good enough glass to take advantage of the difference.
Yeah, but this one goes to eleven.... :-)
Amen.
There is Pixel Qty and Pixel --Quality--.
I use a D1H for any vacation snapshot things and 2.7mp is just fine. I worry not of how large to make a good exposure and 1200 images do not take much space.
I also have a 1Ds 11mp for more serious work and will maybe trade that for a 22-28mp MkIII when available.
Oh, and my wife's Canon Elf at about 3mp - Ideal.
Jon ...
.. and there's a good chance your 10 megapixel sensor is noisier than your 8 megapixel sensor, so a big chunk of that 12% increase in resolution won't survive the noise filter.
random enconters in camera shop...
Why would any one want to pay $XXX for a camera?
That one looks nicer than that other one.
It's got more mega-pixels than the other one, must be better.
random encounters on the street...
One handed shooting; flash in large, dark buildings; small prints and web publishing.
My observations suggests that most people equate high-spec cameras with good photos, mising the whole issue of proper use and good technique.
Shops aren't motivated to educate people when they can easily sell the latest high-tech for big bucks.
"It's almost impossible to see a 12% difference under laboratory conditions with carefully designed test scenes. Realistically, you might just discern the difference between an 8-megapixel and a 10-megapixel image, but probably not."
I agree, Mike. And more importantly to me, in a carefully made print, the difference is even less perceptible than it is on a computer screen with a test image (though it's humorous sometimes how many people still claim they can see a difference). To me, prints are still the real world, not an image in Photoshop at 100% magnification. Instead of engaging in another MP race, I wish the manufacturers would just focus on continually improving image quality in the 6-8 MP range.
But the bug of Konica Minolta Maxxum/Dynax 7D is only able to shoot several thousand photos,
then the shutter is break down.
This is not a good camera on
Sony Alpha 100/KM 7D/5D since
the shutter is too bad.
Hope you can shooting photos over
twenty thousand on your Maxxum 7D.
Take Care. Don't trust your anti-shake from KM7D, 5D/ ALPHA 100: )
Canon had already guarantee to shoot over 100,000. But KM/SONY
is only saying 20,000.
Now the bad shutter of 7D/5D/ALPHA 100 will be a bigger nightmare.
I'm still confused by conflicting advice on this issue. Doesn't it all come down to print size? Yes, a 10 MP camera nominally has 12% more resolution than 8 MP, but isn't that meaningless on an 8x10 print? In fact, wouldn't a 36 MP image look the same as a 6 MP image on an 8x10 print - maybe even worse due to the downsizing?
"I agree, Mike."
John,
Do note that Ctein wrote that post, not me. Although I agree with him too.
--Mike
The difference between a 250 Watt amplifier and a 200 Watt amplifier isn't "a couple of dB", it's 0.9 dB. In other words, it's not a case of "this one goes to 11" but more like "this one goes to 10.1" ;-)
That said, the 12% resolution difference between an 8 megapixel camera and a 10 megapixel camera is just a new variation on an old issue. In the past the question was this: How are people willing pay for a *lens* that has 12% better resolution?
I was quite pleased with the image quality of my D50. It was suberb at high ISO's as well. Got the D80 for the better viewfinder, AF and other features the D50 doesn't have. Not for the extra pixels.
It reminds me a little of something someone said on the job about 15 years ago. We had just upgraded to 486's from 386 class machines (we were a software development shop) and our boss casually asked if we had noticed any speed improvement. The guy at the next desk said, no, but that his machine at home suddenly got slower.
The article seems to be making the case that buyers should be more careful and knowledgeable about what they are buying and should really buy only what is really needed, not what others are trying to sell to us. I can't disagree but we live in a culture where some people have become wealthy by selling cell phones "ring tones". Why do people spend money to make their phone sound different? I haven't got a clue.
What may be happening is that we are so affluent (some of us) that we simply have a need to spend. It's not so important what we buy or why, just that we do so.
In the not-so-distant past, pixel count really did matter. There was a huge difference between my sub-megapixel Sony floppy-disk Mavica and my next camera, a 2 megapixel Olympus. Just as there was a huge difference between my less-than-5 MHz PC and the 8 MHz PC AT.
What we have now is just a continuation of something that used to make sense. But, it will end of its own accord. I'm hearing from friends, who know little about photography, that their files are too big. They are starting to think of a high pixel count as a negative.
What would help a lot is another number, since marketeers and consumers both like numbers, that captures what's really meaningful. Something like "resolving power" or "enlargeability/croppability." The rating can take in more than one measure, something like NFL quarterback ratings (very complicated; maximum is about 158).
Maybe Ctein is the one who can invent the appropriate metric? (It would be awkward if the Leica M8 and the Canon 5D came out in the middle of the pack! ;-) )
--Marc
'photo' wrote:
> But the bug of Konica Minolta
> Maxxum/Dynax 7D is only able to
> shoot several thousand photos,
> then the shutter is break down.
> This is not a good camera on
> Sony Alpha 100/KM 7D/5D since
> the shutter is too bad.
Actually, the shutter in the Dynax 7D/5D is among the best in its class since it's a slow-traveling, low-energy, low-impact shutter that wears out at a slower rate than the competition's fast-traveling high-energy shutters. The drawback of this design is the slow flash-sync speed of 1/160 s (or 1/125 s with AS engaged). The reason is the in-body Anti-Shake system that requires a particularly smooth shutter action. The side-effect is less shutter-induced vibration and thus, sharper images (even with the AS switched off), as well as less wear and tear and thus, a longer shutter life expectation.
'photo' further wrote:
> Canon had already guarantee to
> shoot over 100,000. But KM/SONY
> is only saying 20,000.
Canon---just like any camera manufacturer---actually does not *guarantee* any number of shutter actuations. They only say their shutters are designed for an average shutter life of 100,000 actuations ... that's no guarantee. And Konica-Minolta's shutter in the Dynax 7D/5D is designed for an average 140,000 (!) actuations, not 20,000 (don't know about Sony).
So I am pretty confident that whatever is going to fail on my Dynax 7D first, it won't be the shutter.
-- Olaf
One of the single most absurd and ultimately hilarious comments I ever received on my photography (on a nature photography forum) was that I simply could not take/create a "state-of-the-art" photograph because I was using a so-called "prosumer" digital camera.
Can anyone here tell me just what a "state-of-the-art" photograph is?
Has the fact that I now use (primarily) a high-ish, not the highest, pixel-count dslr elevated my photographs to ""state-of-the-art" or do I still need to go pixel- higher?
This is so obvious I balk at saying it, but is it true that the higher the MPs, the larger a print you can make without upscaling? If so ;-) it seems that pixel count should be a consideration if you wanted to print large?
My observations suggests that most people equate high-spec cameras with good photos, mising the whole issue of proper use and good technique.
Yup. Otherwise amateur photographers would never hear the expression "That's a great picture, what camera did you take that with?"
Fact of the matter is there is a real qualitative difference between certain types of cameras, in certain situations, but (and I really hate to be so elitist about this) anyone asking the above question probably couldn't name a single one of them.
'photo' wrote:
> But the bug of Konica Minolta
> Maxxum/Dynax 7D is only able to
> shoot several thousand photos,
I dunno where you got that notion from, I'm up to PICT8681 on my 7d, no troubles at all. I expect that Olaf Ulrich has the right slant on this -
"Actually, the shutter in the Dynax 7D/5D is among the best in its class since it's a slow-traveling, low-energy, low-impact shutter that wears out at a slower rate than the competition's fast-traveling high-energy shutters"
Consumers just want reasons to spend money.
Megapixels is the easy number (bigger=better, supposedly) for manufacturers to make you feel your 3 month old camera is now trash.
Most consumers really doesn't know what megapixel means by definition.
Regarding shutter life, Canons consumer (Rebel digital) is 15k, Nikons is 15k to 50k (depending on info source).
Forwarded from Ctein:
========
RE: "The difference between a 250 Watt amplifier and a 200 Watt amplifier isn't "a couple of dB", it's 0.9 dB." --Gromit.
Arrgh! Sloppy editing on my part. Originally I compared a 150 W to a 250 W amp. Which is about (pause while doing some mental arithmetic...) 2.2, 2.3 dB ?? Something in there. IOW "a couple."
Then I decided to change 150 W to 200 W to make the ratios more alike in the analogies.Didn't check for dependencies. My bad.
pax / Ctein
========
I saw this article about pixel counting and I got to thinking...
I always like to turn ideas on their heads, and also it can be interesting to pursue the idea of a limit in the mathematical sense. I started to think, 'Well this is all good stuff and sound advice, but what if... you could have an unlimited number of pixels, is there any reason to self-impose a limit? Obviously yes, if the increasing number has a downside such as noise, but suppose noise wasn't a consideration? Well obviously it is but even if it wasn’t then there still must be a minimum number of photons required to be captured to render adequately differentiated 8-bit RGB which again obviously requires a given luminous flux for a given duration for a given photosite size and…
I was about to give up the calculation because it was only a trivial passing interest, when it occurred to me that there was a simpler way of establishing an upper limit to resolution. If the wavelength at the lower limit of human vision is about 750 nanometres then it seems to me that the smallest possible photosites for true colour could not usefully be less than this between centres. If you take a 36x24mm sensor size then you can have (.036 / 750e-9) * (.024 / 750e-9) = 48000 * 32000 = 1536000000 or 1536 Megapixels.
So, panic over. There is no way a 35mm size sensor will ever exceed about 1.5 Gigapixels. Now, perhaps somebody who does this stuff every day would like to demonstrate why the theoretical limit is lower than that and I can stop breathing in this brownpaper bag.