I see this term used all the time in comments about lenses and digital cameras. I’m not aware that anyone has actually managed to clearly define the term, though. Thus, what happens is that one person’s “sharpness” is another person’s “unsharp,” and vice versa.
I love some of the dictionary definitions I found for the word sharpness. For example: "The product or result of being sharp.” Uh, what, like knives? Another says “associated with resolution and contrast.” Okay, are those the only things that define sharpness, and is it just an association or a result?
Most people tend to associate sharpness with a phrase that can be defined reasonably well: edge acuity. If we perceive a clean, well-defined edge that has high contrast, we say that an image has strong acuity. Thing is, other items than the lens can impact edge acuity. Noise (and it’s removal via noise reduction) can destroy edge acuity. Diffraction can lower edge acuity. So is the lens then not sharp? ;~)
Some folk will say “resolution,” or the ability to resolve smaller details is what makes a lens sharp. Of course, the sampling frequency of the digital sensor (i.e. megapixels) might have something to say about this, as does the presence of an antialiasing filter over the image sensor, plus so does the Bayer demosaic algorithm that you use (bet you didn’t think of that last one).
A lens can, and usually does, resolve more than the sensor can actually distinguish. Certainly at its center. These days we have sensors without an AA filter, so we also get false data that looks like “resolving of details” beyond the Nyquist frequency, but is actually faux detail. So is the presence of faux detail okay in determining whether something is sharp or not?
Worse still, many who talk about the “resolving ability of a lens” almost immediately point to MTF tests. But MTF tests measure contrast as a proxy for resolution. Technically, you can have two lenses that can “resolve" the same low level line pairs, but one has better contrast in doing so than the other.
Contrast also plays into edge acuity, too. When we use most sharpening techniques on our images, what we’re actually doing is moving the contrast on either side of the edge so that it is higher than recorded. So do we say a lens is sharp if the camera or post processing can make it look sharp?
DxO, meanwhile, has come up with “Sharpness” as meaning “megapixels.” Did you know that the reasonably well-regarded Nikkor 10-100mm CX lens is only 3 megapixels sharp, but a Nikkor 200mm f/2 FX lens is 34 megapixels sharp? Whoa, wait, what? Get ready for mumbo jumbo:
"Sharpness is a subjective quality…[and our scores are] based on the Perceptual Megapixel concept that weights MTF…with human visual acuity.” Let’s see, that’s a subjective thing that is capable of generating an objective number for a population that has variance in it? How’s that work?
It’s worse than that. Every one of DxOMark’s numbers is based on a particular sensor coupled to the lens, which varies from brand to brand, format to format, and even over time within brand. Even worse, DxOMark makes blithe statements such as “confirms that…”a 12mp full format camera is sharper than an 18mp APS format camera.” Well, okay, the way they’ve calculated their numbers for one particular lens suggests that, but I’m not convinced of the statement they’re trying to confirm let alone their method of confirming it. Further, while they cite what they based their method on, they actually don’t publish the method itself so that we can test the assumptions within it.
I couldn’t figure out why some of DxOMark’s numbers agreed with my personal assessments of various Nikkors, while others didn’t. Then it hit me: test charts. Remember that reference to MTF? Yep, the numbers DxOMark are generating probably devolve to flat field testing at closer distances. What happens with lenses that do better at long distances or that have field curvature?
A good example of the former comes in their positioning of the 200-400mm f/4 above the 500mm f/4 in “sharpness.” The 200-400mm is a sharpness monster close in, but progressively worse as you move outwards towards infinity. This introduces another variable in our task: sharp at what? Close/Far, flat/natural, what?
Then there’s the issue of astigmatisms and coma. One reason why corner “sharpness” often looks poor compared to central “sharpness” is that the lens has clear, uncorrected issues in terms of rendering from side to side. Edge acuity might still be good in the corner, but there’s still something wrong: for instance we get an elongated smear instead of a point (coma is clearly present in that case).
I had a long discussion with my optometrist a few years back. One thing I had noticed is that I have to be careful about whether I evaluate something with my left eye, my right eye, or both eyes. Why? Because my left and right astigmatisms are different. I was asking my eye doc if there was something that we could do to fix this. Since he was one of the first to use a Wavefront approach to lasik-type surgeries, he said sure, but he was convinced that I couldn’t know what my astigmatisms were. In fact, he was certain that I didn’t know what astigmatisms really look like.
So he had me draw the astigmatisms I thought I had and then put me on the Wavefront machine to measure my eyes (he was so interested in my contention that he didn’t charge me for any of what worked out to be a four hour consultation with quite a bit of extra measurement on very expensive equipment). When he got the printout from the machine, he was incredulous: it pretty much matched what I had drawn.
Why is this important? Because when I’m shooting with my camera, my right eye is the only really active one, and it has an astigmatism that’s not fully corrected by eyeglasses that distorts the XY space I’m looking at. If I were to evaluate a lens or my focus on what I see through the viewfinder without considering the astigmatism, it might be slightly wrong. But when I look at images on my computer, I’m using both eyes, and both astigmatisms come into play, even though they’re mostly corrected by glasses (and I wear special glasses corrected for my working distance to the monitor, as well). What I was looking for was a way to remove my astigmatisms from those that my camera/lens were capturing.
No such thing truly exists, though if you get Wavefront or similar eye surgery, you can certainly eradicate many of your astigmatisms, though you may lose some contrast in doing so.
And thus you have one of the things that is contentious about sharpness: it’s not one thing, it’s a combination of things. But more important:
Sharpness is a subjective perception, not a measurable attribute.
The “most sharp” image will usually have additional resolution and contrast compared to the “less sharp” image. Edge acuity will typically be higher on the “most sharp” image. Whether you’re viewing the image optimally will have an impact on whether it looks sharp, too (ask your optometrist at what distance they maximized your prescription for, you might be surprised).
Indeed, you’ll see me and others looking at images closer and farther away when we’re trying to evaluate them. I’ll sometimes take my glasses off and get very close, then put them on and look from the distance they’re corrected for. Why? I’m trying to assess all the low level things that contribute to sharpness correctly based upon my knowledge of how my eyes work.
Which brings me to another point. Most prime lenses have longitudinal chromatic aberration. That’s the kind where you get slightly purple fringing on black and white details in front of the focus plane and slightly green fringing in back. When I’m Autofocus Fine Tuning lenses I find that looking at the edge acuity I see in the results tends to give me slightly false information. Part of that is that depth of field is not 50/50 in front of and in back of the focus plane. But longitudinal chromatic aberration never lies. If you have equal purple/green values at equal points fore and aft of the focus plane, the lens is probably fine tuned correctly.
So, what is sharp?
Technically, a lens is perceived as most sharp if it: (a) has few or no astigmatisms; (b) can resolve fine detail, and at or above the sampling frequency of the capture; (c) resolves said detail with high contrast; (d) renders edges cleanly.
How do we correct for less than maximally sharp? We clean up edges and add contrast to them (proper noise reduction and sharpening in post processing).
Bear all this in mind as you read reviews and hear comments about how “sharp” a lens is. My sharp might not be your sharp. Or it may be sharper ;~).
