Image stabilization

[ErikKaffehr]

Hi,

I think that this discussion is partly without merits. Whatever stabilisation is used I think that the adjustments will be minor. To achieve sharpness a point in the image should be kept steady on the sensor within a few pixels. Moving the image a few pixels is probably all you need for achieving the three stop advantage that most IS systems claim. There are some advantages to lens based stabilisation. Any rotation during exposure will probably around the center of mass of the "camera plus lens" system, and I think that gyrosensors in the lens may better detect this than a sensor in the camera.

Camera based techniques have the advantage that:

a) They work with all lenses
 They are only working when neede (that is during exposure)

Lens based technologies have at least the following advantages:

a) Technology can be optimized for each lens
 The effect of IS visible in the viewfinder


Lens based technologies have at least the following disadvantages:

a) Movable elements and gyrosensors needed in each lens
 May compromoise quality, especially for complex and inexpensive designs.

Best regards

Erik

What focal length range would work well with an in-camera stabilizer?

Within this range, how does in-lens and in-camera stabilizers compare?
[a href=\"index.php?act=findpost&pid=92890\"][{POST_SNAPBACK}][/a]

[gsinos]

I'm just wondering why this is being treated as such an either or decision.  Why would you not want both?  

You would get the advantage of the in-camera mechanism on any and all lenses that do not have the feature.

Certainly Canon and Nikon would be smart enough to have the body disable the feature when an IS/VR lens was attached.

Frankly, I think the big two got caught with their pants down on this one.

One of my friends with a Canon point & shoot (the S3, I think) is thinking of moving up to an SLR.  He wants a body with IS similar to that in his current camera.  

So, he's left with the choice of Pentax and Sony.

Those are the customers that will drive Canon and Nikon to build the feature into the body.  My guess it will be in, at least, the entry level bodies by Christmas 2007.

See you later, gs

[Ray]

Camera based techniques have the advantage that:

a) They work with all lenses
 They are only working when neede (that is during exposure)

[a href=\"index.php?act=findpost&pid=92898\"][{POST_SNAPBACK}][/a]

If it's true that camera based ant-shake sensors only work during exposure (I didn't know that), then it would seem logical that the 2 systems could be combined for maximum benefit.

[howiesmith]

If it's true that camera based ant-shake sensors only work during exposure (I didn't know that), then it would seem logical that the 2 systems could be combined for maximum benefit.
[a href=\"index.php?act=findpost&pid=92943\"][{POST_SNAPBACK}][/a]

I don't think so.  IS senses acceleration and tries to compensate.  Anti-dhake does too,  The one system would not se what the other is doing and the result may be no better than no IS at all.

To work together, the systems would have to detect just the image motion, not camera motion.

[Ray]

I don't think so.  IS senses acceleration and tries to compensate.  Anti-dhake does too,  The one system would not se what the other is doing and the result may be no better than no IS at all.

To work together, the systems would have to detect just the image motion, not camera motion.
[a href=\"index.php?act=findpost&pid=92946\"][{POST_SNAPBACK}][/a]

I shall defer to your expertise as an engineer on this issue, Howie, especially considering the season. Happy New Year   .

[howiesmith]

I shall defer to your expertise as an engineer on this issue, Howie, especially considering the season. Happy New Year   .
[a href=\"index.php?act=findpost&pid=92953\"][{POST_SNAPBACK}][/a]

Don't take my word for it.  Thanks for the holidsy wish.

[jani]

I don't think so.  IS senses acceleration and tries to compensate.  Anti-dhake does too,  The one system would not se what the other is doing and the result may be no better than no IS at all.

To work together, the systems would have to detect just the image motion, not camera motion.

It should be possible to do a bit of both.

That is, allow the lens to do its own IS/VR, and then use a "live" sensor to detect how much movement there is in the projected picture.

But I suppose that this is so computing intensitive that we won't see it before 2010; the power consumption and size of hardware capable of doing this quickly enough is probably prohibitive still.

And even then, I'm not certain it wouldn't suck.

(And yes, happy new year!)

[howiesmith]

It should be possible to do a bit of both.

That is, allow the lens to do its own IS/VR, and then use a "live" sensor to detect how much movement there is in the projected picture.

But I suppose that this is so computing intensitive that we won't see it before 2010; the power consumption and size of hardware capable of doing this quickly enough is probably prohibitive still.

And even then, I'm not certain it wouldn't suck.

(And yes, happy new year!)
[a href=\"index.php?act=findpost&pid=93115\"][{POST_SNAPBACK}][/a]

Not possible without feedback from the image sensor, as you suggest.  Then how does that camera tell the difference beteen camera motion and subject motion (for feedback)?  The subject motion could be in the same direction as the camera's motion, the oppositedirction , or neither?  Which would the sensor follow?

[jani]

Not possible without feedback from the image sensor, as you suggest.  Then how does that camera tell the difference beteen camera motion and subject motion (for feedback)?

There are a number of ways to solve this, and one of the most tempting solutions is to use the algorithms specified in MPEG-4 (if I recall correctly, they may have been there with MPEG-2).

The subject motion could be in the same direction as the camera's motion, the oppositedirction , or neither?  Which would the sensor follow?

That could be configurable.

[gkramer]

If it's true that camera based ant-shake sensors only work during exposure (I didn't know that), then it would seem logical that the 2 systems could be combined for maximum benefit.
[{POST_SNAPBACK}][/a]

The "anti-shake" sensor-based image stabilization sytem doesn't affect the view through the viewfinder, and only affects what appears on the sensor after the image is taken (however long it was "working" before the shutter was released).

Nikon's in-lens VR system (and Canon's as well, I presume) also stabilizes the view through the viewfinder when the shutter button is half-pressed, which is very handy for judging focus, fine-tuning composition, etc. with a long telephoto lens (just as image stabilization is very useful for binoculars of 8x or more maginification, which are virtually impossible to hand-hold). When the shutter button is fully pressed to take the image, however, then [according to Nikon] the "the algorithm changes to compensate for every slight movement [and] just before exposure, the VR lens will reset to central position (optical axis) from an off-centered position which is a result of VR operation during the shutter release button is lightly pressed. Since the shift amount of the VR lens is limited, this operation maximizes VR effects as well as optical performance." Nikon has a useful article (with animations) on their VR system at [a href=\"http://www.nikon.co.jp/main/eng/portfolio/about/technology/nikon_technology/vr_e/index.htm]http://www.nikon.co.jp/main/eng/portfolio/.../vr_e/index.htm[/url]

I'm dubious that combining a well-designed stabilized lens with an "anti-shake" body would be an improvement, and might well degrade the image, as the two feedback systems might fight each other, or alternatively reinforce each other and produce some interesting explosive (overcompensated0 dynamics. But there is a way to find out: Sigma makes a couple of stabilized lenses for various mounts, so one could probably be mounted on a Maxxum-mount (or Sony) DSLR with the anti-shake feature. Anyone tried it?

[howiesmith]

I'm dubious that combining a well-designed stabilized lens with an "anti-shake" body would be an improvement, and might well degrade the image, as the two feedback systems might fight each other, or alternatively reinforce each other and produce some interesting explosive (overcompensated0 dynamics.

[a href=\"index.php?act=findpost&pid=93464\"][{POST_SNAPBACK}][/a]

The link provided appears to answer one question speculated about in this thread.

"Image blur caused by camera shake generally occurs with shutter speeds slower than 1/[focal length] in 35mm format equivalent."

Two independant open loop anti-shake systems can compete with each other, giving variable results (from better to worse than either single system) depending on many external inputs (frequencies of vibrations, amplitudes, directions, etc.)

Another thought about tripods.  As was mentioned earlier, the inertia of the camera system can be easily modified, perhaps by adding weight, or putting the camera on a tripod.  Even holding the camera more or less firmly will change the inertia.  Because the anti-shake systems provide no feedback about relating anti-shake out put to input (actual shake), how does the camera's algorhythm change?  Someone mentioned Canon's lenses do this by detecting the system is attached to a tripod.  But it has no way of knowing how good the tripod is, or how firmly the camera is being held.  It seems only feedback between the sensor (just how stabilized is the image) and the anti-shake system will solve the problem, not another added on antis-shake system.

It does seem possible to apply different programs in the computer for multiple systems, but that is just a band-aide.

Again, this is just my opinion.

[BJL]

The "anti-shake" sensor-based image stabilization sytem doesn't affect the view through the viewfinder ...
Nikon's in-lens VR system (and Canon's as well, I presume) also stabilizes the view through the viewfinder
[a href=\"index.php?act=findpost&pid=93464\"][{POST_SNAPBACK}][/a]

True, except that sensor stabilization does stabilize the VF image when that is done by video preview (LCD or EVF), as in fixed lens "AS" cameras. This could soon happen in DSLRs with a video preview option: most likely first from Olympus, which is already offering the two ingredients but not yet in the same camera: video preview in DSLRs and sensor based stabilization in fixed lens cameras.

[KAP]

I shoot mostly from Aircraft, the results I get from the Canon70-200mm IS are much better than the non stabilised Canon lenses I use. I will also often have a gyro bolted to the camera, in my mind you can't have to much stabilisation, I wish all lenses were stabilised.

Kevin.

[gkramer]

True, except that sensor stabilization does stabilize the VF image when that is done by video preview (LCD or EVF), as in fixed lens "AS" cameras. This could soon happen in DSLRs with a video preview option: most likely first from Olympus, which is already offering the two ingredients but not yet in the same camera: video preview in DSLRs and sensor based stabilization in fixed lens cameras.
[a href=\"index.php?act=findpost&pid=93482\"][{POST_SNAPBACK}][/a]

Quite right, I had in mind a conventional DSLR, not the "electronic viewfinder" point-n-shooters or new video-preview DSLRs (which have so far gotten mixed reviews). Another point (that I found on another blog) is that the AF is not stabilized with the "anti-shake" system, which makes it all the harder to get accurate focus with a long telephoto.

[jani]

Nikon has a useful article (with animations) on their VR system at http://www.nikon.co.jp/main/eng/portfolio/.../vr_e/index.htm

gkramer, thanks for the link.

What surprised me, was the following:

What's more, just before exposure, the VR lens will reset to central position (optical axis) from an off-centered position which is a result of VR operation during the shutter release button is lightly pressed. Since the shift amount of the VR lens is limited, this operation maximizes VR effects as well as optical performance. Only Nikon has this "Centering Before Exposure" feature. (Fig. 3)

I suppose this has been patented, then.

It seems like a reasonable feature to have, since it ought to minimize the needed VR movement at the time of exposure. But it's such an obvious solution that I'm surprised that neither Canon nor other manufacturers haven't come up with the idea themselves, and/or have prior art.

This also reminds me that I lust for that lovely 200mm f/2.0 VR lens.  *sigh*

[BJL]

Another point (that I found on another blog) is that the AF is not stabilized with the "anti-shake" system, which makes it all the harder to get accurate focus with a long telephoto.
[a href=\"index.php?act=findpost&pid=93497\"][{POST_SNAPBACK}][/a]

An excellent point that I had not read or thought of before.
More than ever, it makes me think that sensor-based stabilization is best used only with video viewfinders, not traditional optical reflex viewfinders, meaning in either fixed lens digicams or the "EVIL" cameras that FourThirds is dabbling with.

[howiesmith]

An excellent point that I had not read or thought of before.
More than ever, it makes me think that sensor-based stabilization is best used only with video viewfinders, not traditional optical reflex viewfinders, meaning in either fixed lens digicams or the "EVIL" cameras that FourThirds is dabbling with.
[a href=\"index.php?act=findpost&pid=93652\"][{POST_SNAPBACK}][/a]

The back and forth (near to far) motion that affects focus distance is likely (and hopefully) very very small compared to the focus distance.  No correction required.  Up and down and side to side motion do not affect focus distance.  No corretcion needed.

If up/down and sidedways motion could cause a change in the focused "subject."  Also undesirable.  No coorection wanted.

Maybe anti-shake works OK.

[BJL]

If up/down and sidedways motion could cause a change in the focused "subject."  Also undesirable.  No coorection wanted.
[a href=\"index.php?act=findpost&pid=93680\"][{POST_SNAPBACK}][/a]

I do not understand: for example, if camera shake means that what the central AF sensor is seeing wobbles between the main subject I am aiming at and something in the background, the sensor is going to have a hard time doing its job. (With long telephoto wildlife shots, my main subject tends to be near the center of the frame not off in a "rule of thirds" position). Correction wanted, by me anyway.

[howiesmith]

I do not understand: for example, if camera shake means that what the central AF sensor is seeing wobbles between the main subject I am aiming at and something in the background, the sensor is going to have a hard time doing its job. (With long telephoto wildlife shots, my main subject tends to be near the center of the frame not off in a "rule of thirds" position). Correction wanted, by me anyway.
[a href=\"index.php?act=findpost&pid=93737\"][{POST_SNAPBACK}][/a]

I was assuming, perhaps too generally, that shake would be very small compared to the subject size, and that shake would not be great enough to cause a real change in focus distance (the subject would not move so much that a very near or far point would come "into focus").  Sorry I made this assumption.

It seems the case you have in mind might also cause problems with focus of non-shaking images - if the central subject is smaller than the focus point.

[BJL]

I was assuming, perhaps too generally, that shake would be very small compared to the subject size.
[a href=\"index.php?act=findpost&pid=93744\"][{POST_SNAPBACK}][/a]

My experience with hand-holding 200mm in 4/3" format (FOV like 400mm in 35mm format) is that the VF image can bobble around quite a lot. I have had fairly good success getting sharp hand-held images at low shutter speeds with more normal focal lengths is fairly steady, but those bobbling VF images have shattered my hubris, and make me desire "VF image stability".