| Pixel counts and pixel sizes |
[Mar. 8th, 2008|06:56 am] |
| [ | Current Location |
| | Home | ] |
| [ | mood |
| |  calm | ] |
There are some interesting thoughts about pixel counts and pixel sizes in Michael Reichmann's PMA 2008 thoughts.
While Michael's overall thoughts are definitely right on target, I'm afraid that his numbers might be a bit off, in a way that makes some potentially significant differences.
Michael properly recognizes that the pixel size is an important indicator of how a camera will behave, and that pixels that are too small are ridiculously useless. What he seems to have lost track of is the fact that pixels in point-and-shoots have become really tiny, much smaller in fact than he postulates. He says that point-and-shoots might now have pixels as small as 2.8 microns, maybe even smaller. In fact it's been a while since pixels became smaller than that: the batch of 2/3" 8MP cameras that shipped 4 years ago already had pixels that measured 2.7 microns, an in fact Reichmann himself wrote about that in his review of the Sony F828.
The reality is that point-and-shoots have shrunk their pixels from 2.7 microns about 4 years ago down to 2.4 microns, then 2.2, then 2 microns, and now (with the latest batch) 1.8 microns is quite common (some even go smaller).
Michael is correct when he says that such tiny pixels will cause severe issues with noise: for all I know sensors are still limited to storing at most 1000 to 1200 electrons per square micron. With pixels of 1.8 microns, that means that each pixel is unlikely to hold more than 3900 electrons when fully saturated. Let's assume that there's a 5-stop contrast in a given scene (an evenly lit scene with normal whites and normal black objects, e.g. a bride in a white dress and a groom in a black tuxedo) which is actually not bad at all, and that the camera will be able to meter to exactly bring the sensor to saturation without overexposing, the dark areas will still only get about 120 electrons per pixel (and it's not obvious that colors will be accurate in the highlights when shooting so close to saturation). With 120 electrons to work with, the signal to noise ratio will be about 11:1, i.e. 21dB, i.e. there'll be quite some noise in the dark areas.
If that sounds low, you need to realize that I was talking about ideal conditions: in real-world scenes, light isn't quite even, and in fact it's not uncommon to see a range of 7 or 8 stops across a scene (a bright object in sunlight next to a dark object in the shade will do just that), and sometimes you'll have to bump the sensitivity. I won't even talk about a worst-case scenario with such a camera, but each stop of contrast and each stop of sensitivity costs 3dB in the dark areas, so you can easily figure out where this is going.
Worse than that, those pixels are so small that the lenses that are typically associated with them can't possibly create sharp details (and by that, I mean "details so sharp that the resulting image would be visibly less sharp if there were fewer pixels on the sensor). I have a rule of thumb about that: a lens and a sensor are reasonably well-matched when the peak sharpness of the lens is reached at a numerical aperture that's approximately equal to the pixel size in microns. Let's take the example of the recently announced Sony W300: pixels are 1.8 microns, lens is f/2.8 - f/5.5. Even if we assume that the lens is perfect wide open, that lens would probably be a good match for a 6MP sensor, and that's only for the wide end - by the time you use the zoom, the image quality will degrade even further. To be honest, with such a camera, in spite of is 13.6MP output, I wouldn't bother capturing images larger than 8MP at the wide end and 3MP at the long end. OK, so storage is cheap and capturing 13MP images doesn't cost much, that's true. Still I wouldn't expect the resulting pictures to hold enough quality to print much beyond 8x10 when captured at the wide end, and 4x6 at the long end. Of course, that depends a lot on subject matter, and some subjects might be just fine when printed at larger resolutions, but in general everything will lack texture when captured with such cameras.
On the other hand, Michael Reichmann doesn't think that there's a need to go to pixels much below 5 microns for DSLRs. In general I'd agree with him based on the aforementioned rule of thumb: few lenses peak below f/5. That being said, there are today DSLRs available with 5 micron sensors (the 10MP 4/3 ones), so the notion of having such pixels isn't outlandish. Michael also suggests 60MP medium-format sensors, whose pixels would be 5.4 microns across, and it's reasonable to assume that a 35mm lens can be slightly sharper than a MF lens (since where are fewer design constraints). Michael also has his numbers wrong, and a full-frame 35mm sensor with pixels of 5 microns would have 34.5MP, and not the 25MP that Michael mentions. I'll go even further, and I'll say that pushing down to 4 microns wouldn't be outlandish, since there are some lenses that are so good that f/4 is a reasonable shooting aperture: even Canon's cheap ($300) 50/1.4 would do just fine at this aperture, and all the Canon L primes between 24 and 135mm should be able to produce very good images on pixels as small as 4 microns. Seeing how camera manufacturers have no second thought pushing cameras with pixels much smaller than the lenses can resolve, I'll disagree with Michael here and I'll say that we will eventually see full-frame 35mm DSLRs with pixel counts of 54MP or more. The real question is to know what to do with such images. We're talking about massive prints here, 24x36" at 250 dpi, 40x60" at 150 dpi. Since I currently print 18x24" at 160 dpi, I can really say that such print sizes aren't a stretch at all. The real problem is the cost of printing such sizes: in a nutshell, a printer capable of printing 40" wide costs $5000, and having such prints made in a print shop will take you down $200 per print (or more).
All that being said, I'm still happy with my Canon 5D. It's generally well matched to my lenses and my printer, and while I'd be happy with a few more pixels there isn't anything else available today at a price point that would satisfy me. Until I see a successor to the 5D with about twice the pixel count, I don't have any reason to consider a replacement (and even then I'd end up having to upgrade the printer, and at that size the printer costs as much as a camera, though in that case I might end up skipping one size and jumping directly to an even larger format). |
|
|
![[info]](http://p-stat.livejournal.com/img/userinfo.gif){kind=small}
jbq's journal
