Megapixels - The mythbusters edition

Since the launch of the D800, the internet is ablaze with several discussions regarding the "Megapixel race". Most of them have to do with concerns about noise levels. Every blog post, every forum thread, hell; every tweet has a response from an anti-megapixel pundit who explains why more megapixels are bad for us. Let's take a moment and examine these arguments in detail and see if there's any truth in them.

More megapixels = More noise, right?

Well, yes and no. On a per-pixel level, yes. More megapixels would equal to more noise. This is pretty easy to verify as the first thing most experts on the internet do after downloading image samples from a brand new camera is to view them at 100% on a monitor. But here's the problem.

There's an intended use for every image that one shoots under normal circumstances.

1) To publish on the web, to be viewed on monitors at normal magnification levels, viewed about a foot or so from the monitor.

This involves serious downsizing of the image. Downsizing by itself is an effective form of noise reduction. No matter what algorithm you use to downsize, this involves a reduction in the total number of pixels. When one reduces the total number of pixels in an image, the total number of pixels that have "Noise data" in them are reduced as well. At web viewing sizes, the noise may even be a complete non-factor.

To demonstrate this, let's take a full size, ISO 800 image sample from another camera criticized for having "Too many megapixels", the Sony NEX-7. Let me show you a little crop from the picture that I want to show you. (Picture courtesy: http://www.flickr.com/photos/qwz/6174974652/ )

Nex7_100_percent_crop-1_medium

via i226.photobucket.com


That's a 100% crop from a 24MP, ISO800 image from the NEX-7. It's enough to give most online experts a cardiac arrest in terms of "Noise". It probably would, because this is how they (And sadly, even some "professional" review sites)examine noise levels. Does this mean anything to you in the real world? No.

Consider this, you're contributing to an online pop-culture website that commissions you to shoot pictures ONLY for online use. The same image, the WHOLE image at typical online resolution would look like this:

Nex7_640px_medium

via i226.photobucket.com

Now tell me, do you see the "Horrible Chroma noise" that all the experts cry about? How long did it take for you to realize which part of the image the crop was taken from?

Exactly. Noise is something that has to be seen in context. You cannot judge the real life appearance of an image taken with ANY camera based on 100% crops. Download the sample, use it for your intended purpose and see if it fits. Which brings us to the other intended use for photographs.

2) To print at various sizes.

At normal print sizes, the logic explained above still holds true. When you downsample for printing at say, 6x4, the noise that you see at 100% on a monitor again becomes a non factor. Then there's the "Other" kind of printing. Giant sized, 30" plus prints. Before we get to that, let's do a simple experiment.

Do you have a big HDTV in your home? Have you ever tried standing half a foot away from it and looking at it? Sure, you can see the individual pixels, but it's kinda pointless, innit?

The same logic applies for big prints. When one prints images at 30"+, they are meant to be seen from a few feet away. Download the original image from the Flickr link posted above. At 300 DPI, the original image can be printed at about 20" width. If you have photoshop installed on your computer , open the image in it, set the magnification to print size (If you don't, set magnification to 24% in whichever program that you have. That's roughly the print size for the image) and look at the image from say, 3 feet away from your monitor . Do you see any noise?

I'm sure some of you scrambled right back to your computers to let me know that yes, you in fact DO see "Some" noise. Not as much as when you were looking at it from right in front of the monitor, but you can still make out that it's not a "Clean" image. But here's the deal.

You are looking at the image from the desired viewing distance, but on a monitor that's meant to display things at 96DPI, even though I had saved that JPEG at 300DPI. Pixels on a monitor can't magically "Shrink" to accommodate that. And that's the folly of the mighty internet "Noise expert". For years, we've been told by these guys who've never made a print in their lives that the image "Looks noisy even at print resolution". The reality is that the actual pixels int he print are much smaller than your monitor pixels, so at the same viewing distance, the print image will appear to be cleaner than the one on the monitor!

So hey, if printing large and viewing from a distance makes an image look cleaner, why not print from a camera that's cleaner at 100? Wouldn't that give an EVEN CLEANER print image?

Good question. But there's a problem with that. As mentioned earlier, the NEX-7's sample can be printed 20" wide at 300DPI. The 16MP image from say, a Nikon D7000 (Widely regarded as a very "clean" camera within the same sensor size class) can only do a 16" print at 300DPI. TO get a 20" print from a D7000, you'd have to print it at 240DPI. This may not seem like much, but the larger your prints get, the more the apparent loss of microdetail there will be. Simply put, the image from the higher MP camera may be a tad noisier, but it will also look way more detailed at the same viewing distance. This is why landscapers/ fine art guys prefer cameras like a D3X or a 1Ds over a D3S or 1D.

So I took two pictures, one from a 24MP NEX-7, one from a 24MP A900 and printed them out at 20". Why is the NEX-7 image more noisy?

Again, good question. This is simply due to the size of the sensor. At the same aperture and the shutterspeed, the 24MP Full Frame sensor of the A900 collects more than twice the light as the 24MP APS-C sensor of the NEX-7. More light = less noise. Noise should always be compared within the same class of sensor size. A 12MP cameraphone sensor would have no chance against a D3S (Let's put aside optics for now), but if you make 12" prints from a 12MP cameraphone vs a 5MP one and view them from the appropriate distance, the difference would be a lot less than you thought it'd be. You may even appreciate the extra detail in the 12MP shot.

So wait, are manufacturers conning us with the super high-ISO cameras like the D4 and the 1Dx?

To answer that, you have to understand a simple fact about digital camera sensors. They don't have "Different ISOs". Their sensitivity is exactly what their base ISO is, which is usually ISO 100. The extra sensitivity of any digital sensor is achieved via digital boost. One camera may claim to have a "Native" ISO 51k while another may only have a "Native" ISO12k, but the truth is, both these numbers are artificially boosted sensitivities from the native ISO. The more the boost is, the more noise that's introduced. Any "Hi1/Hi2" etc. that you may see in the spec sheets is the point where the manufacturer feels that the books has reached a point where the noise introduced is a bit too high for their liking.

But make no mistake, your camera doesn't have a toggle within that sets the sensor to ISO100/200/400/800/1600 etc. It's sensitive at ONLY ISO100 natively and everything else is artificially boosted.

Which brings us to the answer to the original question. Cameras like the D4 and the 1Dx have sensors and circuitry within that's more optimized to minimize the noise introduced when the ISO is artificially boosted, for as high a ceiling as possible. High MP sensors like those in the D800 and the NEX-7 on the other hand are more tuned to deliver better dynamic range at the lower ISOs. There's is no "One size fits all" option. If you are the type that photographs black cats in coal mines, boosting the sensitivity of a high MP sensor would result in a flood of noise, one that can't be hidden even by downscaling. For these applications, you need a "High ISO champion" camera. Simple.

Ok, I got all that. But won't my existing lenses be horrible on these high MP cameras? Don't I need to buy "High MP approved lenses?

The answer is in two parts:

1) If you own an excellent piece of glass like say, the Nikon 14-24 f2.8, it will deliver excellent detail on the D800, the kind of which you've never seen from a D3S/ D3X image.

2) If you own a mediocre piece of glass like say, the Sigma 20 f1.8, it will deliver an image on the D800 that's NO WORSE than on a lower MP camera. In the best case, the image would be slightly better, but the advantage would be greatly diminished compared to the above example.

I am not even remotely close to being an optics expert to give detailed commentary on why this is so, but I'd suggest reading experts such as Thom Hogan on this matter, should you wish to know more.

Conclusion:

Don't be afraid of megapixels. They have their uses, if you know how to use them. A while ago, similar concerns were raised about the "Megapixel race" in medium format, but some of the most exceptional images used in advertising and fashion today are taken with scary high MP medium format backs like the PhaseOne IQ180 (80 MP).


I hope this post prompted you guys to see megapixels in context and not rush to the "It's gonna be too noisy" conclusion every time a high megapixel camera is brought up. If you have any questions, do feel free to post here and I'll try to reply to the best of my abilities.

Thanks!