31 August 2009

Physical Limits in Digital Photography

Both spouse and spawn find my tech geek posts impossibly dull, but I promise that this article on the physics of digital imaging is really exiting. Starting with Nyquist limits, what the author found was . . .

Alright, alright, just a few (very geeky) bullet points:

1. The author says that as you crank up the ISO on your camera, the sensor becomes more sensitive (to both light and noise), but you also capture a narrower range of dark-to-light contrast, because "dynamic range drops nearly as fast as the sensitivity increases." Even more interesting, "most of the increase in sensitivity comes from boosting the highlights and midrange: the darkest shadows rendered at low sensitivity are almost as dark." This means that in boosting ISO you both lose dynamic range, increase noise, and gain little in shadow detail. But you also gain little in shadow detail by turning down ISO. What you get in shadow detail is largely ISO independent.

2. "Most [small-sensor DSLR] cameras have lower dynamic range than their [full-frame DSLR] competitors, so the signal-to-noise ratios in the darkest shadows at lower ISO will be slightly worse than full frame, but for the higher speeds they will be up to √2 worse." This would be for, say, a 10mp small-sensor DSLR as compared to a 21mp full-frame DSLR. For a 15mp small-sensor DSLR, it will be worse by a corresponding factor.

3. Compact cameras capture "only 1/25th the number of photons" for equivalent sensor density as compared to a full-frame DSLR. "Most cameras of that size use 8 to 12 or 15 megapixels, so the number of photons per pixel would be about 1/10 the level calculated above [for a 24mp DSLR]." So "we can see why pocket cameras may not even offer ISO 1600 and have visible problems with noise at 400 and even lower." This problem with massively reduced sensitivity and concomitant increase in noise is offset in part by reducing a compact camera's dynamic range. This "lower dynamic range [reduces] how noisy the overall picture will look. But it does not mitigate the problem in the darkest shadows: pocket cameras [need] 10 times brighter light than full frame [DSLR cameras]" to equal their light sensitivity and, therefore, shadow detail.

4. The maximum resolution at which full-frame DSLRs can resolve detail with modern lenses at f10 (their diffraction limit), without degradation, is 21-24mp. For small-sensor DSLRs, at f8 (their diffraction limit), it's about 10mp. This is generally known. Since Canon is releasing a new 18mp small-sensor DSLR tomorrow, there are a lot of skeptics wondering about image quality.

5. Finally, the most interesting bit to me. He says that 35mm film can resolve "close to or perhaps beyond 24 megapixels full frame. But it resolves such detail at very low contrast. At more modest spatial frequencies, film is noticeably worse than digital in MTF/contrast. The net effect is that the perceived crossover point where digital looks sharper than film is in the range of ~3-6 megapixels . . . After scanning some 18,000 slides, I found that my very sharpest slides, taken in contrasty sunlight on Kodachrome 25, are comparable to digital shots at about 6-10 megapixels, at least in terms of resolution (but they are worse in rendering contrast at these resolutions). But for more typical subjects, the crossover point is closer to 3-4 megapixels or less. And for higher speed film or for lower-contrast subjects, it is at 2 megapixels or even less."

He goes on to suggest that this "may account for the fact that the sudden transition from film to digital was largely unforeseen. Photographers were waiting for the ultimate resolution of digital to surpass film and unwilling to recognize that digital pictures looked sharper long before that threshold was reached. One review of one of the first 3 megapixel cameras noted as early as 2000 that the results surpassed those of Velvia film in quality."

The review he cites, by Michael Reichmann, was very controversial when it came out, but it simply points to observable visual evidence in actual photos. No physics or test charts.

There is much more there, but bottom line: A 10mp DSLR beats a 35mm film camera to pieces in terms of effective resolution and sharpness, and at least equals 6x4.5 medium-format film as well. And a latest-gen 24mp full-frame DSLR beats pretty much anything film. And for more reasons than resolution alone.

Want an example? This was shot in a dim arena at ISO2000, no flash, straight out of a Nikon D3x. Try that with film.

No comments: