Monday, January 28, 2008

About camera stabilization

Bert wrote:
... your opinion on in-body vs in-lens image stabilization? The former seems to make so much more sense, both in terms of cost as well as overall weight/bulk, that I wonder why it appears to be the latter which is more common?

Good question.
As readers may be aware, cameras/lenses with image stabilization allows for hand-held photography under darker conditions, meaning longer shutter times, without the pictures becoming shaken.

I've not seen any tests as to what is the more efficient. But both kinds seem to be really good these days.

Actually more brands now have in-body stabilization in their digital cameras (with exchangable lenses). Like Pentax, Sony, and Olympus. But the two biggest names, Canon and Nikon, so far stick with stabilized lenses only. Those two of course say it's because it's more efficient, but they would say that, wouldn't they? One wonders if it's just to protect sales of their legacy stabilized lenses? (Note both brands started making stabilized lenses back in the film days. Film cameras can't have in-body stabilization.)

I wish it were different. Sure, stabilization is more pressing with long lenses, but stabilization gives (with static subjects) a two to four stops gain in practical light sensitivity with any lens, and that's huge. Just look at how much you pay in money and weight to get a lens two stops faster!

It is clear though that stabilization is fast getting cheaper and lighter. Canon has the new kit zoom lens, which is only 200 grams, under $200, and sharp. So it seems to me that they would have no reason to not build it into all new lenses they introduce, even wide angle lenses. Even so my gut tells me they won't, though I can't say why. Boneheaded conservatism, maybe?

Sure, with long lenses it's a minor advantage that in-lens stabilization also stabilizes your viewfinder image, but I don't consider that a deal-breaker compared to the option of having your camera shake-free no matter which lens you put on it. For instance it makes it possible to combine long shutter times with very fast lenses, which obviously is an unbeatable combination for low light.

13 comments:

Mike Johnston said...

You make a nice point about Canon IS lenses being originally for film, but profitability is probably also important. Comparing the prices of IS and non-IS Canon lenses, it becomes clear that they're making a lot of money on in-lens IS.

Another point your post brings up is that seeing the stabilization in the viewfinder might just as easily be considered a drawback as an advantage, because you're not able to detect if the camera is moving or unstable. I'd never thought of that before.

Anonymous said...

I prefer the "stabilized view" with telephotos, as it gives me a better visual sense of the net motion, and helps me correlate the picture results with the visceral experience. With a wider lens, by the time IS matters, the viewfinder blackout time is long enough I lose that sense.

Anonymous said...

"to not build it into all new lenses they introduce, even wide angle lenses"

I didn't do the math, but intuition tells me that building an IS wide lens can be trickier than it would appear at first. The lens groups are so close to each other that any displacement of a floating element is bound to have an enormous effect on the image. It may be tricky to build actuators that precise. You guys, who know the different vendor's offerings quite well, may be able to identify the cutoff point...

"it becomes clear that they're making a lot of money on in-lens IS"

That sure is what it smells like, from a technology standpoint.

The viewfinder stabilization is a good observation though, the kind that would fuel many fiery debates!

Thanks for the insight.

Alex said...

So is in lens IS an active or passive system? Do they use springs and pistons for damping motion, or do they use active to track, which would need some electronics, such a accelerometer or visual sensor?

I should check it out on how stuff works, but thought I'd just ask all you experts.

Anonymous said...

"So is in lens IS an active or passive system?"

To my knowledge, all current IS systems (internal, that is) for still photography are active, of the electronic variety.

They rely on solid-state accelerometer data, and use electromagnetic actuators to move either a floating lens or the sensor itself.

It is basically the same type of actuator technology that is used inside an optical drive to follow the spiral track (concentricity error compensation), etc. Basically, battery-eating demons.

Eolake Stobblehouse said...

What astounds me is how they can move FAST enough to compensate in time!

Anonymous said...

"What astounds me is how they can move FAST enough to compensate in time!"

'Tis not that hard, you know. The only limit on the speed of electro-magnetic actuators is the amount of power you are willing to sacrifice to achieve a given performance level. Just consider the needle speed on an old-style VU-meter. Or the mechanism that positions the heads inside a hard disk drive. And with today's electronics, mechanical vibrations are no challenge at all (sooo slooow ;-) ).

The real tricky part is (accurately) detecting the motion, using a robust sensor. While the IS concept has been around for a long time, it had to wait for solid-state accelerometers to be developed before any really sturdy & miniature systems could be envisioned.

Anonymous said...

This topic has been on my mind for a few days now, and there are a few factors that appear to seriously tip the scale in favor of in-lens stabilization, after all. Note that I am, as often, speaking from an engineering standpoint.

As the focal length increases, the effect of a small camera movement on the image increases proportionally. Therefore, the amount of correction required for a given vibration pattern is also dependent on the focal length of the attached lens.

This has two implications. First and most obvious, the IS system must be aware of the focal length of the lens in use at any given time. While this is certainly feasible, it does imply that, for in-body IS, zoom lenses be designed as to measure and return to the camera the current focal length setting, which somehow voids the claim that such a system can be efficient with any legacy lens.

Pushing this reasoning further, if one system is to achieve hair-splitting accuracy, then the lens must also contain electronic calibration data. Without this, any imperfections in the mechanical components of the lens cannot be compensated for.

In short, zoom lenses designed to work in conjunction with high-end in-body IS systems must contain some electronics, along with corresponding connectors, etc. This somewhat reduces the potential savings that in-body IS would suggest.

The second implication is more subtle. The fact that an in-body IS system is split across two separate and generally unmatched components (the body and the lens) requires that both the camera and the lens be calibrated according to exacting standards. For example, should the gain component of the IS system (in the body) lose its calibration, then the system performance would degrade for all lenses. On the other hand, an improperly calibrated lens would perform poorly even with a properly calibrated body. Strictly speaking, having two distinct calibration operations involved increases the difficulty of achieving a high level of performance, as the two calibration errors (there is always some degree of error) compound each other.

In-lens IS systems present a tremendous advantage on that respect, as any required calibration lies within the lens only and therefore can be performed with utmost precision at the factory. Should a lens lose its calibration, as a result of physical abuse for example, then only that specific lens's performance would degrade, and the body would remain perfectly usable with any other properly operating lens.

There is one final consideration that is worth mentioning. An in-body IS system must be designed to operate across the entire range of focal lengths usable with the system. In other words, it must be able to produce the minute variations required for wide lenses just as well as the larger, faster displacements associated with telephoto lenses. Such a broad spectrum is often hard to achieve, and even harder to optimize.

In contrast, the design of an in-lens system can be tailored and optimized specifically for any given lens. The design of the precision actuators required by wide lenses is very likely to be quite different from that of the actuators found in a long focal length assembly. In my professional opinion, this alone would make in-lens IS the system of choice for high-end performance.

Eolake Stobblehouse said...

"The fact that an in-body IS system is split across two separate and generally unmatched components (the body and the lens)"

What do you mean? The whole system is in the body.

Eolake Stobblehouse said...

"An in-body IS system must be designed to operate across the entire range of focal lengths usable with the system."

Mmm, yeah, but one of my lenses is a stabilized 18-200mm! And the system seems to work fine over the whole of that long range.

Funny, BTW, me speaking for in-body systems, seeing as I keep buying stabilized lenses and no bodies. But if I were starting from scratch, I'd rethink that for sure.

Anonymous said...

"What do you mean? The whole system is in the body."

I guess I wasn't very clear, perhaps I tried too hard to be concise.

What I am saying is that a proper analysis of an IS system has to take into account all of the components that affect the system. This must therefore include whatever mechanism is used to assert the focal length of the lens. In the case of a zoom, there must be something that reports the current lens setting to the in-body IS system. Be it a mechanical linkage, a simple potentiometer arrangement, or a full-fledged, factory calibrated electronic sub-system, at least some part of the system is in the lens.

In my (admittedly hair-splitting) analysis, I seriously consider only the electronic approach as a viable means to achieve high performance.

So, when I say "split across two separate and generally unmatched components", I simply refer to the fact that the body and the lens are manufactured as two separate entities, and are not matched for calibration purposes.

Assuming that the body and the lens each have a 1% error margin on calibration, the net result is a system having a global error margin of 2%, since the errors are compounded (multiplied).

This would not be the case with an in-lens system, because the whole thing is calibrated at once. That is, assuming that the vibration sensor is also located in the lens.

Pls understand that I am by no means implying that it is not possible to build a good in-body IS system. What I am saying is that, as sensor resolutions keep increasing, it will get harder and harder to design such universally adaptable systems. In-lens designs are not trivial either, but it should be easier to attain high performance levels with those.

I may not be an expert in digital photography, but I certainly am at home with electronics. :-) When I read that Nikon recommends turning off IS for tripod use, this tells me that the noise floor (parasitic vibration level) of the IS system is not zero, and not even negligible. Another way to phrase this is that the IS system effectively limits, when in use, the resolution achievable with the lens.

This will limit the usefulness of IS with the newer, higher-resolution sensors, thus prompting for the design of better IS systems. In this perspective, if I had to take a bet, I definitely would bet on in-lens systems.

"Mmm, yeah, but one of my lenses is a stabilized 18-200mm! And the system seems to work fine over the whole of that long range."

Sure, I never said that it was impossible. Now, I may lack hands-on experience with the latest gear, but as a general rule one does not expect the same crispness of image (for example) from a wide range zoom as compared to a prime lens.

While it is apparently quite possible to design an IS system offering adequate performance for such a wide zoom range, there is quite a leap in assuming that the very same system would offer satisfactory performance if used with a high-quality prime...

Eolake Stobblehouse said...

"When I read that Nikon recommends turning off IS for tripod use, this tells me that the noise floor (parasitic vibration level) of the IS system is not zero, and not even negligible."

You could be right. But they might also recommend it just to save your battery. (I haven't seen tests on this point.)

Anonymous said...

"You could be right. But they might also recommend it just to save your battery."

I tried to find where I read the information that led to my comment on tripod use, to no avail. The Internet is such a huge place!

In any case, the recommendation was clearly in the sense of "Turn off IS to avoid ill-effects". Note that it may have been in Canon literature instead of Nikon's, I honestly am not sure.

The one thing I will never doubt, though, is that perfection is not of this world. And while I'm at it, here's another one I like a lot: "Engineering is an exercise in compromise." ;-)