Last Notes on the Megapixel Myth

Last Notes on the Megapixel Myth

My column in The Times last week, about the general irrelevance of megapixels in consumer cameras, stirred up an enormous volume of feedback.

Most of the 700 e-mail responses, including many from experts and engineers, agreed that megapixel count has been promoted (by camera stores and camera companies) as a misleading marketing gimmick for years. My argument—and my test—maintains that 5 or 6 megapixels is plenty for enlargements up to, say, 16-by-24 inch posters.

There were still a few naysayers who don’t believe the results. This one was typical: “Why do you insist in calling it a pixel myth? Other things being equal, 8 megapixels DOES give a higher quality than 5; you just can’t see it.”

(Um, dude? If you can’t see any difference, then it’s not higher quality.)

In this e-column, I’d like to expound upon a couple of points that my print column didn’t have room for.

First, one reason WHY a few more megapixels don’t produce any visual differences in the print.

Let me tease you first with this question: How much bigger can I print a 10-megapixel photo than a 5-megapixel photo?

Most people answer, “twice as big” or even “four times as big.”

But they’re wrong. In fact, doubling the megapixels of a photo actually lets you add only 20 percent more area to each edge.

Here’s the math. A 5-megapixel photo might measure 1944 x 2592 pixels. When printed at, say, 180 dots per inch, that’s about 11 by 14 inches.

A 10-megapixel photo (2736 x 3648 pixels), meanwhile, yields a 180-dpi print that’s about 15 by 20 inches—under three inches more on each margin!

Upping the resolution by even smaller amounts (from 5 to 7 megapixels, for example) produces an even tinier difference—too small to bother with.

[UPDATE: It’s true that a 10-mp shot has twice the TOTAL AREA of a 5-mp shot. But that does not mean twice as TALL a print, or twice as WIDE a print. The “doubled” area translates to only a small additional margin all the way around. My point: the layman assumes that doubling the megapixels means, for example, the ability to make 4-foot-tall enlargements instead of 2-foot-tall ones.]

Meanwhile, cramming more pixels onto a camera’s sensor can actually LOWER the quality of the photo. A former Kodak manager wrote to explain it this way: “Too many megapixels can actually impair a camera’s performance. For example, the typical sensor in a consumer camera is 0.5-0.7 inches. The more millions of pixels, the smaller each pixel must be—and the smaller the pixel, the less light-gathering efficiency it has, and the worse the camera performs in low-light or stop-action shots.”

Now then: At the end of the column, I pointed out that megapixels have become a handy crutch for consumers shopping for a camera. They’d come to rely on it as a letter grade that made comparison shopping easy—and camera makers exploited this fact. “Well, heck, this one has 10 megapixels! That’s a lot better than this 6-megapixel model!”

But a million factors are far more important than the megapixels. The question is: Can we come up with a new one-digit crutch? Can we propose a more meaningful comparison factor?

Lots of you said yes, the sensor size is far more important. After all, it’s undisputed that a 6-megapixel Nikon D40 digital S.L.R. takes better pictures than a 10-megapixel shirt-pocket camera, because its sensor is relatively gigantic. Its individual pixel sensors can be larger and soak in more light, even if there are fewer of them.

Unfortunately, the camera makers and salespeople aren’t going to help you out here. You’re not going to see starbursts in the ads saying, “3/4-INCH SENSOR”! But you should.

In fact, the industry seems to go out of its way to prevent you from knowing what the sensor sizes actually are. It reports digital S.L.R. sensor dimensions in millimeters, like 23.6 x 15.8 mm.

No problem so far. But consumer cameras’ sensors, meanwhile, are reported as a ridiculous fraction, like 1/1.8″—and that’s the *diagonal* measurement. Not only does that mean you have to do a lot of math in your head, but it’s also counterintuitive. The measurements with a bigger denominator actually represent *smaller* sensors. A 1/2.5″ sensor is actually smaller than a 1/1.8″ sensor.

And how are you supposed to compare that to a 23.6 x 15.8 mm digital S.L.R. sensor? Only Einstein knows.

If you can do the math—you can find sensor sizes reported at camera-review Web sites, like steves-digicams.com, dpreview.com, and dcresource.com—you’ll be well rewarded. There are a million factors to consider when you buy a camera, but this one’s a fairly good predictor of picture quality.

A better one might actually be on the horizon. I also received this intriguing message:

“Hi David. I am the author of Imatest software, which is used for measuring sharpness and image quality by imaging-resource.com, DigitalCameraInfo.com, and CNET, as well as many print publications.

“You asked: ‘So what replaces it [the megapixel statistic]? What other handy comparison grade is there?’

“Imatest measures system sharpness as Spatial Frequency Response (SFR), which is pretty much the same thing as Modulation Transfer Function (MTF). These geeky technical terms have great value to engineers, but they scare off consumers, and they don’t quite answer the question, ‘How sharp does an image look?’

“I recently added a measurement to Imatest that does, but it’s unfamiliar, even to most camera reviewers. It’s called SQF (Subjective Quality Factor). It includes print height, viewing distance, and the contrast sensitivity of the human eye. It was used internally by Kodak and Polaroid for years, and it is the basis for Popular Photography’s lens tests—but it was tedious to measure until I added it to Imatest. See www.imatest.com/docs/sqf.html.

“Incidentally, where the megapixel myth really goes berserk is with cameraphones. Limited real estate forces the camera modules to be tiny, which means that the pixels get tiny–well under 2 microns–when the marketing people force the engineers to increase the pixel counts, because megapixels sell. Unfortunately, tiny pixels are noisy, work poorly in low light, and may not be utilized due to a physical phenomenon called lens diffraction. Engineers are well aware of the problems, and they keep butting heads with the marketing people. Guess who has the real power?

“An industry group, I3A, www.i3a.org (worth checking out), is working to come up with better measurements to rectify this situation. It will be quite a battle.”

Comments are no longer being accepted.

A 20% increase in print dimension from doubling megapixels? It should be 40%, since doubling the pixel count gives you a sqrt(2) increase in the number of pixels per edge.

-dms

Pogue’s “megapixels don’t mean what you think they mean” soapbox is probably my biggest gripe with his otherwise great tech column. He’s making a big stink about a non-issue, and possibly confusing the very tech-illiterate people he’s trying to evangelize into thinking that megapixels have no utility at all.

More megapixels equals more liberty to crop. That’s their utility. I hardly think the camera companies are saying to themselves, “Let’s trick people into thinking that more megapixels equals better prints, even though it doesn’t, and even though they have a perfectly legitimate other utility that we will not hype at all.”

I really appreciate this further exploration of this issue. Considering the size of the camera market, and that we all want to record some of our local and personal history, this is a subject that touches all (well, most) of us.

The consumer needs a convenient – and appropriate – metric to sort through the volume of offerings. So, is SQF it? Will steves-digicams.com, dpreview.com, and dcresource.com use it? (Do they read the NY Times?)

Pogue’s Megapixel Malign misses the point for me because his argument includes a premise that I don’t accept: For him it all comes down to the print, he compares cameras based on how those camera’s images will print.

I don’t print! I want the best quality images I can possibly produce, period! I email my pictures, post to a blog, burn dvd slideshows (with zooming, panning & croping), display on my hdtv, I even have people sit with me at my computer and look at my images. And that is just what I do with my images today, in 2007. I am aware that my images almost always have a higher resolution that the monitor is able to display, but who knows what capabilities and options we will have in 2017.

My point is that we need to judge image quality based on the quality of the image, not the quality of the print. Printing is so last century…

David,
very interesting article indeed. You forgot however to mention what is to me the best benefit from larger sensors: narrower depth-of-field.
Sensor size has a similar effect on depth-of-field then aperture.
That is you can achieve bokeh effects (like focusing a face in a portrait and having the background blurred).
I think aperture is also another very important factor. The lower the number the better your lens in low light, regardless of your sensor.

Consumers really enjoy bokeh effects, so I think this could be a selling point from clever marketing departments

#4 wrote: “I don’t print! I want the best quality images I can possibly produce, period! I email my pictures, post to a blog, burn dvd slideshows (with zooming, panning & croping), display on my hdtv, I even have people sit with me at my computer and look at my images.”

Then for YOU, megapixels mean even LESS!!

As you said, a 2-mp camera’s photos would be more than adequate. It would fill up even the biggest HDTV screen with pixels.

–Pogue

David, with due respect, your geometry skills need a bit of work. You claim that people who say you can print a 10-megapixel shot twice as big as a 5-MP shot are “wrong,” and you make your case by saying that less than three inches are added to each edge. But the 5-MP print measures 11×14 (actually, 10.8×14.4), for an area of 156 square inches, while the 10-MP print is 15.2×20.3 or 308 square inches, which is, um, twice as big. The fact that the width and the height are not twice as long is irrelevant. It’s the area that matters.

Hasselblad are completely transparent on this issue. On all their marketing stuff about, say, the H3D-31 it clearly says 31 Mp on a 33x44mm sensor. If you’re serious about cameras, I don’t see what you have to complain about.

Great article, David. More tech journalists should take it upon themselves to educate the suckers… er, I mean consumers… who believe all the advertising hype.

Image noise is unfortunately a property of most small sensors and gets worse as pixel density increases. The problem is by no means limited to camera phones. Most of the popular point-and-shoot digital cameras suffer from excessive noise at higher ISO settings, especially in low light (which is precisely where one would crank up the ISO setting). This one factor (noise) has prevented many an otherwise excellent camera from earning a “highly recommended” rating from DP Review. Camera makers do sell “more pixels” and are without a doubt selling their customers short in the process.

Also note that another extremely important factor in overall image quality is in-camera image processing. Canon is the industry leader on this one. Nikon is nearly as good as everybody else is lagging behind.

Cheers.

i like that david keeps going on about this, but i do agree that print is maybe not the best demonstration of the problem here. most people aren’t printing large pictures so it’s somewhat irrelevant to me if you can or can’t see the difference between a 5 or 8 megapixel camera in print.

i think the real problem is that the consumer is led to believe that they’re getting a better product by buying a camera with more megapixels when in fact a camera with less megapixels may actually perform better. i doubt any of us would be as accepting of this marketing gimmick if let’s say intel were to sell us more expensive processors that performed slower than less expensive older processors.

i don’t buy the argument that the real freedom afforded by a higher pixel count is the ability to crop. how many of the average consumers out there are actually even using even basic photo editing tools like crop and red eye correction? i’m skeptical that it’s many. also, what exactly is the value in cropping when the end image you’re left with has more chromatic and luminous noise than should be acceptable?

i suggest that if you’re really doing that much cropping that there’s another great tool you can use: your feet.

David,
if I remember right, these diagonal inch dimensions originate from ancient imaging sensors (a so called vidicon). These vidicons are glass tubes. The outer diagonal of this tube is the referenced value, but the photosensitive area is much smaller. That means if you have a 1″ CCD the usable diagonal would be around 16.4mm and not 25.4mm. The same logic applies to 1/1.8″ CCD sensors.
That means the sensors we buy are even smaller than the manufacturers make us believe!

I would also give my strong opinion on megapixels: My old Canon G3 with 4MPs makes much better pictures than my new super-compact Canon 8MP camaera. Of course it fits better into my pocket.

11 x 14 = 154 square inches of image and 15 by 20 is 300 square inches…looks double to me…perhaps I am not focusing correctly.

Regards,

Greg

Are the lag times of various cameras ever compared? That seems to be an essential consideration.

While the article maybe does make sense for your way of using digital photos it certainly is not a universal truth. Cropping is just one of the reasons for wanting higher resolution than using digital files for 1:1 prints.

In addition you have the relation between file size and print size/area totally wrong. You state:
But they’re wrong. In fact, doubling the megapixels of a photo actually lets you add only 20 percent more area to each edge.
Here’s the math. A 5-megapixel photo might measure 1944 x 2592 pixels. When printed at, say, 180 dots per inch, that’s about 11 by 14 inches.
A 10-megapixel photo (2736 x 3648 pixels), meanwhile, yields a 180-dpi print that’s about 15 by 20 inches—under three inches more on each margin!
The truth is this:
double the pixel count can (with the same print DPI) yield a print that is 1.428 times bigger on each side (and BTW an edge doesn’t have an area) and exactly double the area. Think you need to back to geometry class – this is pretty basic.

Such total error in basic understanding totally undermines your viewpoint – total loos of credence.

I agree that megapixels are overrated and megapixel stats are overused, and used wrongly. One area where most cameras can improve is dynamic range. For example a picture with snow, rocks, and shadows. Is the snow just a white blob and are the shadows all black?, Or can you make out texture on the snow and see what is going on in the shadows. Most pictures from most cameras have blown out highlights and muddy shadows. Fixing this will require sensors that work better, not just sensors that have more megapixels.

By your calculations, going from 5 to 10 megapixels results in the print size going from 11 x 14 (154 square inches) to 15 x 20 (300 square inches). I would consider that very close to twice as big.

I don’t get Steve-O’s comment that “Printing is so last century…” I guess he’d have us walking around with digital viewers strapped to our heads so that we don’t see the real world at all, just a high-tech computer screen. This is a very impractical vision, to say the least… no pun intended.

Steve-O would perhaps be right if the only use for photographs was to display them for the enjoyment of friends. For that, the internet does indeed rule.

David
This is what you said in your column.
“Let me tease you first with this question: How much bigger can I print a 10-megapixel photo than a 5-megapixel photo?

Most people answer, “twice as big” or even “four times as big.”

But they’re wrong. In fact, doubling the megapixels of a photo actually lets you add only 20 percent more area to each edge.

Here’s the math. A 5-megapixel photo might measure 1944 x 2592 pixels. When printed at, say, 180 dots per inch, that’s about 11 by 14 inches.

A 10-megapixel photo (2736 x 3648 pixels), meanwhile, yields a 180-dpi print that’s about 15 by 20 inches-under three inches more on each margin!”

5 mega pixel picture=11×14=154 sq inch print
10 mega pixel picture= 15×20=300 sq inch print

If you compare 154 vs. 300 sq inch print; a 10 mega pixel camera does give you about “twice as big” picture area compared to 5 mega pixel camera.

Well, the stuff about sensor size and cramming pixels is often valid. A more accurate gauge of sharpness is shooting a line count chart. But the glaring math mistakes of the column are absurd: a 15″x20″ print is obviously more than 3″ larger in each dimension than a 11″x14″, who did the proof reading here? And of course doubling the pixels gives you 1.4 more info, as was pointed out earlier. For a tech column I expect better.

Another significant factor affecting image quality is the quality of the lens. A DSLR will typically have a better lens than a pocket digital camera. Combined with a larger sensor the better lens will result in higher quality images. But if all you do is make 4 x 6 prints, nobody will notice…

David,
I agree with your conclusions, but if you are “doing the math” you need to be a little clearer.
You say that the 10mp photo isn’t twice as large as the 5mp, and say it adds “20% more area to each edge.” I think you mean it adds 20% more to each dimension’s length.
In actual area, if your dimensions are correct, the 10mp photo really is about twice as big:
5mp=11″x14″=154 sq. inches
10mp=15″x20″=300 sq. inches
You are getting twice as much photograph, even if it doesn’t look that way. If you doubled the dimensions, the area would quadruple.
Your column is great, and it really is the quality of the pixel that counts, but I had to comment on the math (something I do a lot of).
Thanks,
Scott

Interesting but I have a few points to quibble:

1. “When printed at, say, 180 dots per inch, that’s about 11 by 14 inches.” What printer are you using at 180 dpi? (actually it should be ppi, dpi is an offset printing term) Most minilab printers (for instance, the Fuji Frontier) print at 300ppi for up to 10×15. Larger printers, like the Durst Epsilon, use 254ppi or thereabouts. No lab printer goes down to 180. Now an Epson inkjet printer can use 180, but if your image includes fine detail, you should be using the higher 240 or 300 ppi resolutions. My HP inkjet printers all want 300, and I prefer not to let my printer make image decisions for me.

2. The megapixel war may be promoted by the camera marketing departments, but the stock photo agencies are driving it for professionals. Workbook Stock’s file requirements are 60MB tifs; Getty Images requires minimum 47MB tif files. Even though the vast majority of images won’t be run larger than 25MB (that’s about a full page image, and as you know in the print world, you don’t really need 300ppi for press use anyway), the editors want lots of pixels for the same reason many consumers do: cropping. No matter how tightly or artistically the photographer may have framed that image, some editor wants to be able to screw it up!

It is more than megapixels, but I think you really need to define your argument more carefully with regard to point and shoot cameras versus dSLRs. A lot of people are going to assume that a 6MP P&S is equal to a 6MP dSLR from your arguments, and I think that’s exactly what you don’t mean.

I’m not a great photographer; my composition frequently sucks.
But back at the computer I can crop pictures to get better composition, if there are enough pixels to let me do that and still print a large format image.
There’s a limit, but 5 Mpixels is just a little light. I use 10 these days and it is working really well for me.

I’m curious about David’s choice of 180 dpi for his print size calculations. Magazine editors want 300 dpi, and I’m wondering if this is because of some “dpi loss” inherent in the magazine printing process. I print snapshots at 300, but I’d gladly use 180 if that is usually adequate for uncropped small prints.

Dave,

I’m not a photography expert, but I think that your statement below may be in error:

“In fact, doubling the megapixels of a photo actually lets you add only 20 percent more area to each edge.

Here’s the math. A 5-megapixel photo might measure 1944 x 2592 pixels. When printed at, say, 180 dots per inch, that’s about 11 by 14 inches.

A 10-megapixel photo (2736 x 3648 pixels), meanwhile, yields a 180-dpi print that’s about 15 by 20 inches-under three inches more on each margin!”

I think that the comparison should be done as area, not as a direct comparison of length and width. An 11×14 page has an area of 154 sq. in., and a 15×20 page has an area of 300 sq. in. Therefore, using area, a 10-megapixel photo can display a print about twice as large in area as a 5-megapixel photo.