Focus!

Many years ago I researched the issue of focus. I read about all the various means of determining and setting focus on various types of cameras, focusing aids and how they work, and the issues affecting accuracy and repeatability. I have to admit that I don't remember all of the technical arcana, so if I wanted to explain it all, I would have to go research it again. But learning it once was an eye-opener. It turns out that although focusing a camera is almost universally seen by practicing photographers as easy to do, it's in fact not a trivial technical problem, if we adopt suitably high standards for accuracy. Almost always, what we're doing when we focus our cameras is setting approximate focus. It's only because we're usually focusing on complex, 3-dimensional subjects, and because we're granted considerable leeway by depth-of-field, that approximate focus is usually good enough.

Carl and I just went through an interesting exercise. He had prepared an illustrated addendum to his article showing how he determined and set the custom lens focus adjustment  on his K20D. The problem we had was that when he reduced the test files to web resolution, the fine distinction between in-focus lines and out-of-focus lines on the test chart literally disappeared. What he was seeing at full resolution in ACR was not possible to show at web resolution.

I noticed a different aspect of the same thing all those years ago when I did my research and tests on focus. The greater the resolution of the whole system, I found, the easier it was to detect exactly where the plane of focus was placed, and the more evident it became that objects just out of the d.o.f. were not in focus. Somewhat counter-intuitively, if I reduced the resolution of the whole system (for instance, by using faster, coarser-grained film and hand-holding the camera), it looked like more was in focus. All I was doing was making the highest levels of distinction between the plane of best focus and the closer zones of the d.o.f. disappear. To my eye, at least, this made most pictures look better, rather than worse.

It's easy to demonstrate the principle in Photoshop. Click on the picture above to bring up the larger version, and see if you can tell where the focus was placed.  Not easy, is it? That's because I've applied a liberal amount of Gaussian blur to the entire picture. To my eye, it looks like everything from Vietnam Inc. to maybe Leonard Freed is resolved pretty much the same, and it's not evident where the plane of best focus actually is (except that you might reasonably assume it's in the middle there somewhere).

In this sharpened version that was not blurred, it's much easier to tell where the plane of best focus is—approximately on The History of the Japanese Camera. In this version, it's possible to see that both New York Rises and Lewis Hine are slightly out of focus—something that's invisible in the top picture.

This demonstration doesn't quite make the point, simply because the entire top picture looks obviously blurry. But, in fact, the entire bottom picture is fairly low resolution—it's a small JPEG, 72 ppi, and the camera was handheld with the lens wide open. In it, I can't tell much difference in sharpness between History, Bruce Davidson Portraits, and Wright Morris (the pink book on the far side of History.) Make a really high resolution picture of this subject, and you might even be able to tell which side of the spine of History is more in focus!

When you've achieved enough resolution so that the "sharp parts" look subjectively sharp, then having more of the d.o.f. match the best apparent focus can make more of the picture look sharp. That's why small apertures often make pictures look subjectively sharper even though, in objective terms, diffraction is degrading the sharpness of everything in the frame.      

'Best practice'
But back to focus. Strictly speaking, all methods of focusing have pitfalls and potential for error, and all types of focusing aids have their characteristic limitations. That's even without taking into account the mechanical "slop" and the general imprecision of our devices (and other factors that affect apparent focus, such as film flatness or the flatness of field of our lenses). Once you add that in, it's fortunate indeed that only approximate focus is commonly required, because we'd be out of luck if we had to do better consistently!

As it was, I adopted one 'best practice' that I continue to observe habitually to this day: I don't use a lens's widest aperture at its closest focus distance. If I'm focused as close as the lens will go, I make sure to stop down at least somewhat, and if I'm using the widest aperture, I make sure I'm focused at least a little farther away than the lens's closest focus point. The reason is simple enough: there just aren't very many camera/lens combinations that are both precise enough and accurate enough to focus as close as they can go with the lens wide open. Assuming you'll get good focus in such situations just leads to frustration.

In the case of AF systems, consumer cameras, especially ones that are less than top-of-the-line models, are not terribly high-precision devices (they have to be pretty darned good, all things considered, but they're never perfect). There's a fair amount of slop in the design and construction that has to be overcome. But even so, most AF systems are "detuned" from their best theoretical performance. The reason is that the more accurate the AF and the finer the system's tolerance, the slower the system will achieve focus and the more it will hunt. If the tolerance is widened, speed increases and hunting decreases. The camera-makers know that the typical end-user is simply less aware of precise focus than he is of AF speed and hunting, so they compromise. In the best cameras, not only is the system itself better and more capable, but the balance between focus accuracy and AF speed is skewed more towards accuracy (despite which, the AF still has to be fast). That kind of performance, of course, comes at a cost.

___________________

Mike

Comments

[Mário Nogueira]

How I work: I compose the scene, and then i decide what I want to keep in focus. Using a measuring tape, i get the nearest and the farthest point of focus. I then read a DOF table, and find what aperture and focusing distance to use. I blindly trust on the DOF chart... ;)

Of course, for some subjects (far landscapes) I just use the hyperfocal...

Cheers!

[stevierose]

A small addition to your last paragraph about today's consumer cameras and the compromise between focus accuracy and focusing speed: The issue with most of todays small sensor digicams is that they have (because of the physics of optics) huge depth of field. The "35 mm equivalent view" of your typical "28-105" digicam is really 4.6 to 17.3 mm. It really takes some work with these cameras to set up a situation where you purposefully create a limited DOF for say a portrait with out of focus background. So, of course, the manufacturers concentrate on AF speed and are sloppy about focus accuracy. Most of the time everything is going to look "in focus" whether you want it to or not.

[John Camp]

Something I never thought about until I saw your examples -- there is also a subjective impression of focus that has nothing to do with camera focus, movement or bad technique. It has rather to do with the contrast of the subject. In your second example, both Vietnam Inc. and the History of the Japanese Camera seem to me to be better focused than the white Bruce Davidson book, which is between the other two. That's simply because the type color of "Vietnam" and "Japanese" are in sharper contrast to their backgrounds. I don't know if this means anything, or if I should worry about it, [scratches head] but it's interesting.

JC

[Gordon Lewis]

This posting strikes a chord. I made a similar post (http://shutterfinger.typepad.com/shutterfinger/2008/04/what-you-see-is.html) to my own blog a few days ago. My point was that what you see on your computer monitor is so software-dependent that the only practical way to evaluate how sharp an image is is to print it.

The only thing I'd add to what you've said so well already is that today's autofocus SLRs (film and digital combined) have focusing screens that are tuned for brightness over accuracy. They need to be bright because only part of the light is reflected by the reflex mirror. The rest gets passed through to the AF and metering systems. Unfortunately, the brighter (and smaller) the focusing screen, the less useful it is for accurate manual focusing, which I fear is becoming a lost art anyway.

The conclusion I draw from this is that how "sharp" an image is or how accurate the focus is mostly a matter of how small the margin of error and closely you inspect it. The question is, what's the best way to insure maximum accuracy and sharpness when needed? I suspect that with many cameras, "close" is as close as you can get.

[Andre]

Mike,

I read an article on Luminous Landscape (http://luminous-landscape.com/essays/digital-world.shtml ) awhile back that suggested that the focus point shifts slightly when the lens aperture is changed. Assuming this is true, and given that most (all?) SLRs focus with the lens wide open and then stop down only when the shot is taken, wouldn't that consistently introduce a bit of focus error? I really need to try this with my old screw mount Takumar as it has a manual stop down mechanism that would allow me to focus both wide open and at the shooting aperture.

-Andre

[Stephen Scharf]

I'm beginning to think the best auto-focussing camera I've ever used is my original Canon 1D (Mk I). Now *that's* a camera that can autofocus. I used it during the MotoGP weekend at Laguna Seca when the shutter on my MkII suddenly failed during Saturday morning qualifying, and it delivered absolutely beautiful results. Put a Canon 70-200/2.8 or 300/2.8 on that puppy and you will know what focus precision and accuracy is all about.

http://photos.imageevent.com/puma_cat/motogp2007/ChrisT9.jpg

[David Bostedo]

This is a bit off topic, so I guess feel free to moderate it out of existence if need be, Mike. Anyway, with regard to this sentence :

"...it's a small JPEG, 72 ppi..."

I see this somewhat frequently, and it bugs me every time. A JPEG being set to 72 ppi in no way affects the size of the file, either on the disk or in terms of resolution. A JPEG file that is 800 pixels by 600 pixels is that size no matter what the ppi is set to. Only programs or output devices that use inches (printers, Powerpoint, etc.) will care about the ppi setting, and they will only use it to reconstruct the physical size of the image. It still won't change the resolution.

So I could have a 20 MB, 10000 x 7000 JPEG and 72 ppi, and it would certainly not be a small JPEG.

[Martin]

More than a year ago, I wrote a series of articles (in german) about the issue of AF of DSLR cameras.
Briefly my conclusions (badly translated):

-AF-sensors are usually bigger than marked in the VF

-therefore they easily found a better place to focus on with higher contrast behind or nearby your subject than you wanted

-contrast does matter

- shutter speed - naturally - does matter

-do not use AF-servo if the subject does not move

-recomposing after using the central AF-point for focussing to the side at distances til 5m my lead to out of focus (make a sketch, take a ruler and you see why)

- a lens with f 2.8 or greater focuses by accident within a the range of DOF (at open aperture of the lens); with f2.8 or smaller (f2.0 etc.) the lens focuse by accident in a range of 1/3 of the DOF, ergo probably more exactly.

This is equal to the industrial standards of "what is sharp"
(with our own eyes focussing on a groundglass we could never do better)

-and last but not least: a picture in sRGB may look sharper than in Adobe RGB.

[Mike Johnston]

Andre,
That's called "focus shift" and it can indeed be a problem, although I've only seen it for sure with enlarging lenses. It shouldn't be a characteristic of good modern lenses.

One advantage of a good enlarging grain magnifier if you're doing darkroom printing is that you can focus at the working aperture and focus shift doesn't come into play. Some view camera photographers also focus wide open and then recheck focus at the taking aperture. I'd hate to have to check for focus shift with a reflex viewfinder. I don't think I could tell. With a rangefinder camera, there's no way to tell if you're getting focus shift or not.

Mike J.

[Mike Johnston]

David,
You're right. I do that habitually and I keep forgetting that what you say is true.

Mike J.

[Mike Johnston]

JC,
You're right, but actually my example picture is a little misleading (read: sloppy) because the books aren't all lined up. "Vietnam Inc." is actually pushed quite far in (there are actually two thin books you can't see between it and the magazines you see next to it), so it's actually closer to the plane of focus than the books you see next to it.

Sorry, but I just did those pics very quickly. It's one of the nice things about the web, as opposed to book or magazine illustrations...it can be Q&D....

Mike J.

[Hugh Crawford]

One of the wonderful things about the Sony R1 is that it uses contrast detection on the sensor itself to focus. A little slower , but much more accurate. Most DSLRs basically guess how much the lens needs to be moved to bring the subject into focus, then move the lens without checking, in what is called "open loop" in the software world. I find that actuating the autofocus on a DSLR a few times after initial focus until the lens stops moving is necessary, closing the loop. Even then you are subject to the vagaries of focus shift when the lens stops down, or textures freaking out the DSLR autofocus.
Since the the Sony R1 uses contrast detection to focus at shooting aperture, and you can move the focusing spot to an arbitrary location in the frame, you never have any of these problems. Of course it's slower that the initial guestimation of a DSLR, but it is accurate.

I hope that contrast detection focusing gets included as a feature in some of the live view DSLRs.

[XebastYan]

Mike,
next time use your CD or VINYL Rack as example *gg*

groove on
Xeb.

[william]

Another factor to keep in mind is that the on-thirds/two-thirds rule-of-thumb about how the point-of-focus relates to the DOF is only an assumption. It is reasonable to assume that if you focus on an eye, the nose-to-ear region will be in focus.

Some lenses do not behave this way.

Case one

|------*------------| what we expect

Case 2

|--*----------------| what we might get

Legend: |------------| DOF

* point-of-focus

In case 2, the photographer may assume there is a focus error (I focused on an eye but the face is soft) when in fact the DOF symmetry is responsible.

[Hugh Crawford]

"That's called "focus shift" and it can indeed be a problem, although I've only seen it for sure with enlarging lenses. It shouldn't be a characteristic of good modern lenses."

If I recall correctly focus shift is caused by spherical aberration and the placement of the aperture a little in front or behind of where it should be to prevent focus shift. Of course you can exploit that to move the acceptably sharp DOF to be mostly behind or mostly in front of the plane of critical focus

Nikon produces a 105mm and a 135mm AFD with defocus control where there are two apertures at two points in the lens.

Sony makes a 135mm f/2.8 - f/4.5 STF Smooth Transition Focus lens , but they use an aperture that is not a solid hard edged hole , but a blurry "apodization element" to achieve a visually similar result. The "apodization element" seems to be something like a clear center gradated neutral density filter, sort of the opposite of those filters designed to counteract vignetting in extreme wide angle lenses

I've been wanting to try these lenses for quite a while, a lens that combined these two designs would be pretty interesting too.

The problem with lens testing by the numbers is that a little spherical aberration and focus shift makes for bad test scores but beautiful photographs.

[Ctein]

Dear Mike and Hugh,

I haven't used anything but modern-design, six-element enlarging lenses since my college days, so I can't speak to lesser optics. But I have never seen a focus shift worth mentioning with such top-tier lenses unless I was trying to use them well outside their recommended magnification range (like doing slide duping or making murals).

~ pax \ Ctein
[ please excuse any word salad. MacSpeech in training! ]
======================================
-- Ctein's online Gallery http://ctein.com
-- Digital restorations http://photo-repair.com
======================================

[XebastYan]

I found this one:
"Sharpness is a bourgeois concept"
Henri Cartier-Bresson

[Mike Johnston]

...And, conversely, Oren tells me there are, in fact, many camera lenses known to suffer from focus shift.

Mike J.

[Mike]

Mike,

Apparently the Zeiss ZF 85 1.4 exhibits quite significant focus shift, requiring focusing at the selected aperture up to about f2.8 - even on bodies that will operate auto stop down.

Nice to read about manual fous in this day and age though:)

Mike

[Jason]

This is a very dense read today. Very informative but very dense. My only thought to share here is that I did not know about shooting both wide open and at the shortest focusing distance is not recommended. It makes sense having read the explanation now, and explains why so many of my macro shots from a nature walk I took on Sunday came out soft and/or blurred. Thanks!

[Ctein]

Dear William,

The 1:2 DOF rule is so usually wrong as to be useless. I have no idea where it came from, other than someone's innumerate imagination.

It's not a function of the lens; it's inherent in the equations. At close distances, the DOF is symmetric. As the subject gets further away, the DOF becomes more asymmetric, until the ratio goes to infinity at the hyperfocal distance.

There is a single, intermediate distance at which the ratio happens to be 1:2, but who cares?

pax / Ctein

[Damien]

I use a couple of manual-focus lenses on my Pentax K10D, and quickly realised that the AF confirmation dot was completely insufficient for the task.
So, I got a 10$ focusing screen with split-screen and microprisms off ebay. It works fine ! When focusing on a vertical line with a 50/1.4 lens at the closest distance (45cm) and at full aperture, I get consistent results precise down to 1 or 2mm.

[Scott Kirkpatrick]

Focus shift is especially troubling in well-regarded older fast rangefinder lenses used now in the unforgiving digital environment, where they are judged by 100% enlargements. Zeiss and Cosina/Voightlaender have introduced "classic" lenses recently which sharpen and shift focus as they are stopped down. Field curvature seems to come in the package with these older-style designs. In at least one recent lens (the CV35/1.4) it appears that objects at the edge of the frame are in focus closer to the camera than the objects at the center, and may even be sharper when the lens is wide open.

scott

[Andre]

As a follow up: I tried the grain magnifier with a Schneider 50mm enlarging lens and couldn't convince myself that I could see any focus shift at all. I'm not sure if it's more or less comforting to know that it only affects some lenses. This is shaping up to be one of those things that I wish I had never stumbled across.