Depth of Field—How Far To Stop Down
See also Depth of Field.
This question keeps popping up in online forums.
For the purpose of this discussion, let’s assume we’re discussing the Nikon D2Xs, a camera whose photosites are located 5.5 microns apart (which doesn’t mean all 5.5 X 5.5 microns are light-sensing). The Canon 400D offers similarly tiny photosite spacing (actual sensitive area might be different however). Newer cameras might have larger or smaller photosites.
Circle of confusion
First, depth of field tables assume a 30-micron “circle of confusion” roughly termed a “blur spot”. That’s about 30 times larger than the photosite spacing on a Nikon D2x. Thus, an image from a 12.2-megapixel camera is considered “acceptably sharp” when its resolution is reduced to about 0.4 megapixels. That makes no sense at all, certainly it is no argument for purchasing a D2x! Try downsampling your sharpest D2x image to 0.4 megapixels and then making an 18 X 12 print—how does it look?
Print size is often referred to as one variable in selecting the depth of field required. Unless you really are sure that you will never print larger than W x H, you should ignore the print size argument—otherwise why would you buy any camera that had more resolution than would required for that size? A D2x would be pointless if you were always making 12 X 8-inch prints; a 4 or 6-megapixel model would do just fine. Try it—a Nikon D2h makes stunning 12 X 8 prints. (This assumes the same sensor size; a consumer digicam won’t fare so well due to poor pixel quality).
Sharpness in a print. The required resolution to make a “sharp” print will depend both on subject matter and perceived sharpness, not just measurable resolution.
Why risk the chance of not being able to sell an image simply because it was sub-optimal in terms of its crispness? The client asks for a 30 X 20 and you’re stuck trying to make a soft image look sharp at that size. No fun indeed. For those not selling their images, why would you risk not being able to make a larger print if desired? So shoot each picture for as much quality as you can get at reasonable effort (resolution, contrast, color accuracy, etc). Paying thousands of dollars for a professional or prosumer digicam makes little sense if the images don’t utilize that resolution (unless your needs involved usability and handling issues).
With a head shot (portrait), 5mm of focus error can cause slightly blurred eyes at wide apertures, or one eye in focus and the other out. No calculation needed there--it has to be spot-on. For portraits at wide apertures (f1.2, f1.4, f2), the effect is simply lost if the focus is off. If you want both eyes sharp on a head shot (and they're not both aligned in the same plane as the sensor), you'll probably need f5.6 or f8 or even f11 for head and shoulders images.
Contrast can be as important as resolution, because the human eye perceives contrast strongly, even more so than resolution. This is why “low resolution” images from a Nikon D2H (or similar camera) can look very sharp when printed as large as 30 X 20.
Depth of field calculations don't address contrast well at all. In fact, increased depth of field (smaller aperture) almost always reduces contrast past f8. Contrast is influenced by the particular lens and its coatings and aberrations, and the f-stop and lighting (among other things). Stopping down too far drops the contrast and the image starts to become “flat” (and actually unsharp). Post-processing can help, but it’s no cure-all. MTF (Modulation Transfer Function) graphs show clearly how contrast drops as lenses are stopped down, as well as how it drops at higher resolutions, one reason the D2x sometimes suffers with less than the best optics. See any of the Leica lens pages for examples.
With a few exceptions, f/5.6 is optimal for sharpness and contrast on most DSLRs and most lenses. Throwing depth of field into the equation and f/8 is probably the best all-around choice.
Aperture f11 is generally just as sharp, but there is a slight loss of contrast. At f16 contrast drops substantially. At f22, everything will look blurred, no exceptions, when compared to f8/f11/f16, unless you’re shooting a really, really poor lens.
In the future with 30-50 megapixel full-frame DSLRs, the optimal aperture will drop to f/4, beyon which diffraction will muddy the image. However, one must always achieve a balance between depth of field and other factors.
With the outstanding lenses, stopping down 1 stop can yield 90% of the potential lens quality, with two stops yielding nearly all of it. Examples of this include the 28-70/f2.8 ED-IF AF-S, the 60mm/f2.8D Micro Nikkor, the 85mm/f2.8D PC-Micro-Nikkor, and the 200mm/f2 ED-IF AF-S VR. There are quite a few others. Does this contradict the f8 rule? No, not really—f8 usually shows some minor improvement out to the corners, even with the best lenses, while rarely degrading the center. A truly outstanding (theoretical) lens which has been perfectly assembled can be diffraction limited, meaning that no stopping down is required for optimal performance. This just doesn’t happen with any of the 40 or so lenses that I own. NASA might have some though.
Consumer grade lenses, or even crossover consumer/pro lenses might benefit from stopping down to f11 at some focal lengths. For example, the Nikon 18-200/f3.5-5.6 AFS G VR is preferably shot at f11 at the long end, though f5.6 can be quite good towards the shorter end.
Real depth of field
Real depth of field and theoretical depth of field are not the same. Lenses with chromatic and other uncorrected aberrations won’t produce the crispness that the DOF calculations would predict.
Also, two different lenses at the same focal length might also produce different subjective impressions of depth of field, with one showing out-of-focus color fringes, and the other showing none. Different lens designs simply won’t produce the same results when stopped down. Internal focus (IF) designs can sometimes be inferior for landscape work to their (older) non-IF counterparts when stopped down. I observed the strong preference of Bjorn Rorslett for his older non-internal-focus 85mm Nikkor for precisely this reason (a photographer who has many lenses at his disposal).
If you’re skeptical of all this, that’s just fine with me. It’s best to run your own tests and see for yourself. Just be aware that performing a valid test is no trivial task, starting with getting focus spot-on, which can be quite tricky (see Focus Accuracy).
Note that the strongest image might mean minimizing depth of field, while ensuring precise focus. This is a different effect, but often more compelling than an image with everything “sharp”. Indeed this is the reason so many photos taken with consumer grade lenses are boring—the maximum lens aperture might be f3.5 or f5.6, and as a result pleasing blur is never obtained.
So which f-stop?
The Nikon D2x has a 5.5 micron photosite size. The Nikon D3x is closer to 6 microns. Canon cameras are similar.
Simplifying the discussion to assume there is a 5.5 micron photosite, the camera would become diffraction-limited around f17. In practice, my tests show that f11 is the limit for good micro-contrast while maintaining the resolution seen at f8, with both resolution and contrast dropping quickly after that. You can stop down about 1 stop more for the Canon EOS 5D, and about 1/3 stop more for the Nikon D200 because of their larger photosites.
Rule of thumb
Assuming you want significant depth of field, choose f8 first. Stop down to f11 if more DOF might be beneficial, and use f16 if DOF is really important to your subject. Do not stop down to f22 unless and overall blur to the entire image is acceptable.