Noise Performance Comparisons

At some point I thought that the megapixel race was over when manufacturers started to back off with their insane resolutions. A very good example was the move from the Canon G10 at 14Mp to its successor the G11 at 10Mp. A more recent example is that of the iPhones. Apple was smart enough not to bump up the megapixels but to increase the sensor size of their phones while maintaining resolution. In the case of the Canon G10 to G11 example, the main motivation there was that the G10 produced really ugly noise at ISO 200, just one stop higher from it’s base ISO. That didn’t last long. As soon as sensor makers discovered more ways of improving sensor performance the megapixel race started again at an even more insane pace. Nikon was on pole position with the release of their APS-C sensor Nikon D7000 which was quickly followed by their full frame D800 at a staggering 36Mp (essentially the same D7000 sensor but bigger). Now you can get APS-C sensor cameras at 24Mp from just about any brand. Bigger is better! Or is it?

It’s common knowledge that one of the biggest factors affecting high ISO noise performance is the size of the sensels (sensor pixels). Ultimately it’s the individual sensels that capture light. Bigger sensels mean higher signal-to-noise ratio (SNR) assuming of course that everything else is equal. Technology plays a big role as well. My Olympus E-M5 (m43 @ 16Mp) for example is way cleaner at high ISOs than my older Sony A700 (APS-C @ 12Mp) no matter how you look at it. Yes, older Sony sensors sucked. Big time. Now everyone else is using Sony sensors because of their miraculous performance (thanks Sony). I would like to emphasize this “no-matter-how-you-look-at-it” because this is exactly what I will attempt to explain in this post.

There are always trade-offs every time you alter components of any technology. In the case of sensor noise performance it is a delicate balancing act between sensel size and resolution. Sensor manufacturers face the difficult task of improving sensor efficiency as technology attempts to miniaturize everything (think of mobile phone cameras). For me, it is quite frustrating that as soon as they discover new ways of improving sensor performance they increase the resolution as well. Imagine if the D7000 maintained a 12Mp resolution instead of 16Mp or the D7100 at 12Mp instead of the (stupid) 24Mp. Wouldn’t we be getting ISO 400 performance at 3200 by now? Well maybe not that much of a leap but you get my point. It’s like two steps forward and one step backward. Why?!!!

The saddest thing is that gear heads are still gear heads. They want the latest of everything but they have no choice but to ride with the flow. In the end, the crazy consumers do the justification in behalf of camera manufacturers in the form of heated arguments in forums. A lose-lose situation for photography.

Let me expound on this “justification by consumers”.

Not all of us are gear whores. Some of us actually use our brain. We prefer pixel level performance over resolution. I, for example, prefer a brand new D700 (12Mp) over a D800 (36Mp!!!). Actually I think a D4 at 16Mp is better but it is too expensive and the Df is ergonomically unusable for me (please refer to my quick review of this camera). I might go with 24Mp if they did not downgrade the D700 to the levels of the D610 but 24Mp for me is the maximum I would go for a measly 35mm sensor. So at the moment, I really don’t have much choice but to stick with a very old D700. I will buy another D700 if you can find me a brand new one at a reasonable price.

Increasing sensor resolution is not only detrimental to noise performance but it is also very impractical especially for landscape photographers like me as I have explained in my post on diffraction limits ( I looked at the noise comparisons in dpreview of current and not-so-current full frame cameras and it makes me question whether we are really getting improved image quality with these new cameras. Here’s a screenshot from the high ISO comparison in dpreview:



Do you really want me to believe that the D800 is cleaner than the D700? If you are not anywhere near legally blind, it’s quite obvious that a 7-year old sensor still outperforms the newer cameras in terms of SNR.

It doesn’t stop there. D800 owners perform “tests” of their own. You can see they are cheating because they compare noise performance in good light. I mean, hello? In good light, SNR is obviously higher than in low light. In good light, only the stupid will bump up their ISO. When there isn’t much light, the D700 shines.

You think that’s funny? Wait ’til they pull the next trick out their sleeves: resampling!!! That’s right folks. If you downsample the D800 it will outperform the D700. That’s 36Mp down to 12Mp. It’s easy to see why this “works”. Because noise is random, they tend to cancel out when resampled. This is essentially the same effect (although not quite) with videos. A movie looks cleaner because at 24fps the random profile of noise will ultimately cancel out each other thus leaving behind a clearer “picture”. Try to pause a movie though and you will see how ugly it is. Clever trick isn’t it?

Clever but not really fair and it most cases it doesn’t really make sense. You bought a 36Mp camera so you can reduce it to 12Mp? Admittedly, it is quite nice that you have this option. I mean who prints larger than 12×18 anyway, assuming that you ever print at all? Can you see where this is going? A 12Mp image at 240dpi will produce a print size of 12×18. That’s huge! And that’s why we chose a 12Mp camera. It’s more than big enough. If we do need a larger print then we upsize in Photoshop. Upsizing makes sense for prints. You do not upsize and image just so you could pixel peep deeper. Upsizing for the sake of pixel peeping is stupid because no new data is added. You upsize because you expect a print to be viewed at a farther distance which negates any perceived increase in noise. And that brings me to the counterpart of D800 downsampling and that is D700 upsampling. ROFL!! Of course if you upsize a D700, the tiniest of noise will now look like boulders! If you really want to compare 36Mp with 36Mp then stitch several 12Mp shots of the D700 and compare with the D800. Makes more sense, yes? But they won’t do that because that will expose their stupidity.

I mean, let’s be honest here folks. We are comparing sensor noise performance. Do it fairly. View the images in their intended sizes: that’s 12×18 for the D700 and 20×30 for the D800. Any trickery beyond that is just cheating. Justification by trickery is unacceptable when you are trying to be objective.

In the end, it’s what you do with your camera that matters. Your 36Mp is useless is you won’t even print larger than 4×6 because your shots are not even worth hanging on your wall. Yes, you have the potential but it takes skill and hard work to make that expensive piece of new equipment shine. Bottom line is that all this discussion on noise performance is just a big pile of cow manure. Leave that problem to the engineers. Our problem as photographers is how to wake up early for that perfect shot.

Until then.

5 thoughts on “Noise Performance Comparisons”

  1. But surely the only meaningful way of assessing noise performance, in practice, is with a constant image size? Comparing a tiny image produced with an old phone with a large image produced by a full-frame camera is meaningless, unless image size is the only thing you are really worried about.

    1. The only objective way is by looking at one pixel at a time a time and averaging it. Has got nothing to do with the image. It’s a useless exercise.

  2. Hi, I’m reading your blog since yesterday, and I found it very interesting. Can you tell me, is this true, what this guy ( start from 0:30, he’s my night-timelapse guru) is saying? That you can only shoot stunning night timelapse with FF (because of the larger sensor). Is it possible to capture shots similar to his work with APS-C or m43 (that would be delightful, because I love m43 format, although I don’t have any such camera yet). I believe that camera has to perform very well in low light conditions in order to do it. Are any of m43 cameras capable of compare bigger sensors cameras regarding high ISO? (sorry for my English, it’s not my first language)

    1. A m43 is more than capable of timelapse. Modern sensors like those in the E-P5, E-M5 and E-M10 produce acceptably clean images even at ISO 3200. Couple that with a fast lens such as the 12mm/f2 and you get yourself a low-light monster.
      Avoid the E-P3 and E-M1. Their sensors have horrible noise in long exposures. I’m not familiar with Panasonic m43 cameras but I have heard good things about them.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s