I don’t think it’s a stretch to say that one of the more attractive features that make the full frame VDSLRs attractive is that the large sensor can render shallow depth of field. When compared to a Super 35 frame, you’re looking at almost a stop narrower depth of field for the same aperture. For example, a 50mm f/1.4 lens on a 5D mark III would render a scene about the same as a 35mm f/1 lens when used on a Super 35 sensor like the one in the Canon C100.

So, the big sensor sensor in the 5D mark III is good for thin depth of field. What about when you want lots of depth of field?

My first reaction was that I’d need to consider another platform; something with a smaller sensor that I could leverage for more depth of field. That was, at least, until I started running some numbers, which actually surprised me.

Aside from the limit imposed by the lens itself, diffraction is the biggest problem with stopping down much beyond the diffraction-limited aperture for a sensor. For a 5D mark 3 with 6.25μm pixels would be diffraction limited around f/9.3 as N = (6.25e-6) ÷ (1.22 × 550e-9) = 9.3, at least when shooting stills.

However, when shooting video, the 5D mark III, as best as I can tell, bins 9 pixels in a 3×3 grid into a single pixel for the video output. This is in part of the reasons the 5D mark III has a 5760-pixel wide 22MP sensor instead of a 21MP sensor like the one in the 5D mark II.

In any event, at least somewhat naively, when shooting video the effective pixel you’re using is 3 x 6.25μm or 18.75μm, which put the diffraction-limited aperture at rather lovely f/28. Not at all bad, especially considering that most of my lenses won’t even stop down that far.

Compared to some similarly theoretical numbers form other cameras.

A Canon C300 uses a 4 pixel 2×2 grid (1 red, 1 blue, and 2 separate greens) to produce an RGB tuple video pixel. However, they apparently aren’t demosaiced, instead they’re taken directly to form the RGB tupple. With 6.4-micron pixels, you can expect the C300 to start showing diffraction effects at apertures between f/9.5 and f/19.

Using the worst-case number, and comparable fields of view, a C300 will show softening at the depth of field equivalent aperture of f/18 on the 5D mark III. Using the best-case number, the depth of field equivalent aperture[i] steps out to just over f/25. A C300’s (or C100 for that matter) crop factor doesn’t help if you’re looking for more depth of field.

A Canon 7D (or 60D, or T3i, etc.) uses an 18MP APS-C sensor with 4.3-micron pixels. The readout is considerably more complicated than either the C300 or the 5D mark III as it uses binning and line skipping to produce the resulting video file. The makes the computation of a diffraction-limited aperture significantly more complicated, especially since Canon doesn’t fully qualify how the video is produced. However, I would expect to see diffraction effects at somewhere between f/6.5 and f/13 while shooting video. However, the results will likely be asymmetric due to the way the sensor is read out.

Using the worst-case of f/6.5 the DoF equivalent aperture on the 5D mark III is f/8. With the best case of f/13, the DoF equivalent aperture is f/20. Again, the crop factor doesn’t help get more depth of field, at least not before you should start losing sharpness due to diffraction.

The short of it, is that the 5D mark III’s pixel architecture for shooting video makes it something of an anachronism. It’s seemingly capable of both incredibly shallow depth of field when shot wide open with fast lenses, as well being very capable of stopping down without significant image quality loss when depth of field is actually needed.

Of course stopping down isn’t a perfect solution. Having to stop down further to achieve the same, depth of field may result in raising the gain and as a result potentially produce noisier video. However, that’s a project for another time.