Why isn't autofocus perfect?

[wogears]

<p>Y'see, sonny, back in the days of manual focus, if the distance from lens to film was exactly the same as the distance from lens to screen, then you were in focus, period. (Unless you stopped down significantly, in which case focus shift could occur.) Of course if the camera was out-of-spec, then there were errors, but I'm considering 'best case' here. It seems to me that live-view manual focus on DSLR HAS to be equally 'perfect' (at the focusing aperture) because it is taking data from the sensor itself. AF from the sensor should be the same. (Assuming a subject with reasonable contrast and sufficient magnification.)</p>

<p>This being so, I have seen instances on the Internet where people complaining of back-focus or front-focus have actual rigorous test shots to confirm this problem. (A minority, to be sure--most of this is operator error.) Most significantly, the problem appears to occur with some lenses, but NOT with others (same test conditions). Is there something about phase-detect AF that 'moves' the focus point based on the lens in use? Just what, technically, is going <em>on</em> here? Why is 'focused' not <em>focused</em>?</p>

[sarah_fox]

<p>Great question. I'm curious about that myself.</p>

<p>AF is a process of successive approximation. At some iteration the focus is deemed "close enough," and the AF process is terminated. My hunch is that there are patterns inherent with individual lenses that tend to throw the focus systematically one direction or the other.</p>

<p>Another factor is field curvature and spherical aberration. If you're focusing with Canon's 24-70/2.8, for instance, the focus using light from the outer margins of the lens will focus in a different plane from the light through the center of the lens. This probably relates somewhat to focal distance. As this is a property that varies between lens designs, there may be some consistent focusing errors that need compensation. Of course this would have been true with the old split-prism and microprism focusing systems. Then again, field curvature is more a property of zooms, and we didn't use those quite so much back in the day.</p>

<p>And finally, the AF sensors might be incorrectly positioned, requiring compensation for all lenses attached.</p>

<p>This is all conjecture, of course, aside from the last point. I'd love to hear/read something authoritative on the subject.</p>

[howard_m]

<p>it would seem to me that there are manufacturing tolerances for the lens and flange distance etc etc and those can be cumulative, thus effecting AF accuracy w/ certain lenses.</p>

[Tim_Lookingbill]

<p>I thought this issue was already dealt with and explained.</p>

<p>We don't know if these focus tests shoot under a diffused light and/or use a focus target made up of thin fine lines on white paper which tends to throw off a camera's focus metering. I always get spot on focus points shooting in bright sunlight, using flash and under halogens in a dimly lit museum. But if I try to focus in dim fluorescent lit rooms and shade at dusk/dawn outdoors, it's usually hit or miss.</p>

<p>I have my focus button assigned to the OK button separate from shutter release. Often in dimly lit or low contrast scenes I'll hit the OK button and it seems to sharpen up, hit it again and it micro adjusts without moving the camera. Hit it again and I can hear the focus ring make another micro adjustment. It appears it's sensing something my eyes aren't seeing. Do a zoom on the camera's LCD screen and it's either soft at the point I focused on (AF set to Spot) or it nails it. Edit the image and find it wasn't the focusing that was the problem, it was a clarity/contrast issue. How do you tell the difference out in the field?</p>

<p>This is why I don't trust sharpness tests.</p>

<p>Believe it or not, sometimes I have to reduce sharpness in post on some Raw shots of sunlit scenes with my $80 18-55mm Pentax K100D kit lens. Go figure.</p>

[geoffs1]

<p>Live-view focus should be essentially perfect (in the sense that what you see on the screen when focusing should be identical to what you'll see when you look at the file later).<br /> However, most DSLRs don't use the sensor for focus. They have separate AF sensors that can be mis-aligned just like the focusing screen can be mis-aligned. I would guess that is the source of the common consistent (i.e. all lenses) front or back focus problems that people find.<br /> Here's a link with a nice diagram: <a href="http://www.leongoodman.com/d70focus.html">http://www.leongoodman.com/d70focus.html</a></p>

[luis_g]

<p>Perfection is for the Gods. It was not perfect back in the MF days (except for Leicas). It makes life interesting.</p>

[JDMvW]

<p>I'm also of the opinion (implied) that MF was on the whole less accurate than AF is today.</p>

<p>Try to focus a stop-down f/4.5 lens on a Contax S ground-glass prism and you will see the inherent difficulties easily.</p>

<p>For me, the ideal AF lens is one that clearly marks which sensors are active and which allows manual touchup of the focus without having to switch to manual focus.<br /> Ideal, that is when my eyes aren't acting up.... :( then pure AF is best.</p>

<p>Things are also complicated in that many younger users really don't KNOW how to focus manually. There are web pages teaching that but, because some of the best of them are certain kinds of commercial sites, P.net doesn't like links to them on some forums.</p>

<p>From <em>The Bank Dick</em><br>

<strong><a href="http://www.imdb.com/name/nm0840316/">Og Oggilby</a></strong>: Oh... I knew this would happen! I was a perfect idiot to ever listen to you! <br /> <strong><a href="http://www.imdb.com/name/nm0001211/">Egbert Sousé</a></strong>: You listen to me, Og! There's nothing in this world that is perfect.</p>

[bob_sunley]

<p>One other thing not mentioned is some lenses suffer from focus shift when stopped down.</p>

[stp]

<p>You ask a good question, Les. I have no answer to contribute, but I wish manual focusing were as easy as it was in the days before AF. Those glittering micro-prisms that would suddenly vanish when focus was achieved were very easy to use, especially in conjunction with a good matte screen. Split image finders were also good (at least with non-telephotos) if one could find a relatively thin, elongated object or a sharp edge (e.g., a twig or the edge of a leaf) near where the focus should be. Today, I have to have a lot more "faith" than in decades past when trying to achieve an accurate focus. Live-view is good, but it takes pushing of more buttons to get there, and it's sometimes difficult on a long tele because of camera shake as the manual focus is being adjusted. Like many other aspects of modern digital cameras, focusing is a feature that is still being engineered in the hopes that it will be improved (e.g., the Canon 1Dx).</p>

[wogears]

<p><strong>@Tim, etc: </strong>I am not talking about test errors. I am talking about cases of good testing practices, where the manufacturer <em>acknowledges </em>that this problem can occur with a specific lens while <em>not </em>with others and repairs the lens. And I <em>did</em> mention focus shift on stopping down in the OP.</p>

[peter_langfelder]

<blockquote>

<p>AF is a process of successive approximation.</p>

</blockquote>

<p>Phase detection AF is not a process of successive approximation. That's why phase detection AF is so fast and allows predictive AF. In fact, I suspect (and have read something written by Bob Atkins to this effect concerning Canon EOS cameras) that the camera makes a calculation of optimal focus, sends the lens there and does not look again to check whether the focus is optimal. In fact, it would be difficult to acquire one-shot focus on a moving subject if the camera "checked" focus after the lens arrived at the requested focusing position since the focus would never be good.</p>

[peter_langfelder]

<p>One more thing - it used to be "common knowledge" (again from writing of Bob Atkins at <a href="http://www.bobatkins.com/photography/eosfaq/eosfaq24/9miscellany.html#q27">http://www.bobatkins.com/photography/eosfaq/eosfaq24/9miscellany.html#q27</a>, based on information from Chuck Westfall) that Canon EOS AF worked with circle of confusion of about 0.035mm on film, so you could make an about 8x enlargement and still maintain "critical" focus (i.e., circle of confusion of ~0.2mm) in the final print. This circle of confusion related to DOF scales, but I believe it also related to the AF precision (Canon used to state that the AF will focus "within the depth of field" of the lens at full aperture). The 0.035mm may have been good enough for film, but these days, with sensor pixel spacing of a few microns, it may not be enough - of course unless Canon refined their definition of "depth of field". In any case, I would not be surprised if the AF were not precise enough for today's high-density imaging chips.</p>

[Jerry_]

<p>The lenses for a Hasselblad 500C/M had a scale system. If you had the lens at f11, everything [on the scale] from the close-point to the distant-point was in focus. Rather simple, no electronics needed.<br>

And with a lens like the wide-angle 60mm, <strong>it worked well</strong>.</p>

<p> </p>

[Spearhead]

<blockquote>

<p> everything [on the scale] from the close-point to the distant-point was in focus.</p>

</blockquote>

<p><br />You'd think that with the internet, nonsense myths like this would have ended.</p>

<p>There is exactly one plane of focus, i.e., things can only be in focus at exactly one distance. Everything else, moving forwards and backwards, is progressively out of focus. This is readily visible in prints larger than a contact print.</p>

[Tim_Lookingbill]

<p>Sorry Les, I wasn't aware of this specific problem existing especially the part about it being acknowledged by the manufacturer.</p>

<p>I do have back focus problems but it's intermittent and frustrating as hell figuring out the cause. I always thought it was the AF metering not presented enough contrast and so just hunts for the nearest hard edge it senses, but your issue doesn't seem to apply to that in this case.</p>

<p>Frankly I see so many quirky things going on with my digital camera never mentioned in any of the tests and reviews that sometimes I think I don't have the same model camera or it's just malfunctioning.</p>

[leslie_cheung]

<p>Perfection might be interesting in, say, mathematics. Great arts, like us mortals, are more about limitations, flaws and (often senseless) emotions. </p>

[colin carron]

<p>I still am confused about this. Geoff's link shows how a DSLR focusing system works which explains how a camera body could produce the wrong focus. My experience though is that occasional lenses will not focus correctly while other lenses are fine on the same body. So that implies the fault is in the particular lens - not the body.</p>

 

[robert_k1]

<blockquote>

<p>Why isn't autofocus perfect?</p>

</blockquote>

<p>On a film body, I am used to metering and focusing manually through the viewfinder, often at different locations within an image. Once either the focus or exposure is set, it remains locked, and I can deal with the other.</p>

<p>I find doing so on a dslr very challenging, especially if I auto focus. The problems:</p>

<p>- Dslrs' AF settings are very complex, involving knobs and buttons on the body, and many menu options. Setting the correct permutation of these correctly for a particular shooting situation can be mind boggling. My camera's manual does a poor job clarifying these.</p>

<p>- Exposure metering and focusing are tied together in most cases. It takes some doing to figure out how to separate the two.</p>

<p>- Using live-view to check focus is fine for static subjects, but not for moving subjects.</p>

<p>http://www.photo.net/nikon-camera-forum/00ZVO0</p>

<p> </p>

[dan_south]

I have a handful of autofocus beefs. The first is a flat wall that is exactly parallel to the sensor. AF should be

able to determine how far the wall is from the camera. But if the wall has no contrast, e.g. a seam or a shadow cast

upon it, it's invisible to AF circuits. The lens hunts around but never finds the wall.

 

Problem two happens when objects are very far away. I have a feeling that the AF thinks that all faraway objects exist

at infinity. Infinity for the lens might only be a few yards away. That's not the proper focus for a distant mountain

range. And the result is a fuzzy enlargement.

 

Finally, there are faces. Funny things happen when AF looks at a face, even when I am careful to point the active AF

sensor at the near eye. Sometimes the AF focuses on the collarbone. Sometimes the hairline. Sometimes the

earlobe. Perhaps it's a problem to be solved by chaos theory, but it drives me nuts.

[robert_k1]

<blockquote>

<p>Finally, there are faces. Funny things happen when AF looks at a face, even when I am careful to point the active AF sensor at the near eye. Sometimes the AF focuses on the collarbone. Sometimes the hairline. Sometimes the earlobe. Perhaps it's a problem to be solved by chaos theory, but it drives me nuts.</p>

 

</blockquote>

<p>Agreed with all your beefs, but in particular this one.</p>

<p>On my D200, there are menu options for AF-Area Modes, Focus Zone Selections and Predictive Focus Tracking. On top of these, there are "dynamic-area AF", and "closest subject priority AF". It takes a PHD and endless trial and error to decipher and get the right permutation of settings for each shooting situation.</p>

<p>Take your face focus as an example. I would use Single-Area AF to focus the eyes, and still get the problems you described. In the manual, the area size of Single-Area AF is not defined. Depending on the size of the eye being focused within the viewfinder, the focus area may or may not work well.</p>

[Tim_Lookingbill]

<p>It appears from what Dan and Robert K describe you'ld think there's one camera manufacturer making all the camera's with each company just putting their logo on it.</p>

<p>I get the same problems only I don't have all the options that requires a PHD to fix it. Why don't they just build a better, more sensitive AF metering system? Kodak's not doing anything lately, give it to them. Looks like they need the money.</p>

[robert_k1]

<blockquote>

<p>Why don't they just build a better, more sensitive AF metering system?</p>

 

</blockquote>

<p>I think the problem is not that the dslrs' AF system is not "sensitive". But the manufacturers can certainly provide better explanations on how the AF system work beyond the dictionary like user's manual, and offer some suggestions on different settings for different situations (like recipes). Nikon used to publish such tech notes, etc. on how to use their flashes.</p>

<p>Here's another example of why it takes a PHD to understand AF:</p>

<p>http://www.photo.net/nikon-camera-forum/00ZXGL</p>