Visualizing image stabilization

EclipseLast month I went outside at 3 AM to photograph the eclipse. But I ended up having a hard time getting a good picture of the moon. The pictures were turning out unreasonably blurry– much more than I’d expect from just the moon’s apparent motion during the exposure time. The problem turned out to be the “IS” in my Canon S3 IS– the image stabilization– which apparently needs to be turned off for this sort of thing.

But why? Isn’t image stabilization supposed to take out blur?

Pointing elsewhere in the sky, you can sometimes (depending where you live) see small point-like sources of lights that can provide a useful tool for figuring out what the bleep your camera is actually doing. In the land of moderate light pollution one thing that we can see through our electronic viewfinder is the Pleiades star cluster, so let’s point our lens at that. The parameters for the two photographs below are identical, except that they were taken with image stabilization off and on, respectively: 15 second exposure, aperture wide open, zoomed in like crazy.
No IS.jpg
IS.jpg
In the top photo the stars each look like clean, easily distinguished stripes.

(In order to reduce vibration due to the button press, these two pictures were taken with a timer delay of a few seconds. However, there is still a small, squiggly tail at the base of each star track, presumably due to residual vibration of the tripod.)

In the lower photo, with image stabilization turned on, you can really see a significant difference. The initial squiggly tail at the lower left of each star track is still present, but is now smaller– thanks to the IS no doubt. Otherwise, the shape of the tracks is quite different. Let’s zoom in:

At the lower left is the initial squiggle from the tripod. In the middle is a large almost triangular structure where the star light was initially steered to. Then, there is an additional, wandering shape starting towards the upper right that bends down the right side.The net effect is that the trail of the star is not reduced to a point– just bent around into a loop.

Conclusion? It looks as though the image stabilization works well for short times– maybe up to a second– but lacks accuracy for longer exposures. And *that* is why you might want the IS off to take a clean picture of the moon.

Update: From what I can see online, there’s no image processing component to the image stabilization process– it’s based solely on sensing acceleration of the camera. Possible answer: Could the slow wandering of the image result from noise in the accelerometer signal?

 

13 thoughts on “Visualizing image stabilization

  1. I’m going to guess that the IS is trying to stabilize during the entire time of the exposure and sees the apparent motion of the stars (or Moon) due to the Earth’s rotation as if it were camera motion. Atmospheric interference may also come into play.

    It seems hard to believe that the IS is this "smart" though. I had always assumed that IS in digital cameras was a simple matter of periodic sampling as if the shutter were snapped at regular intervals with the actual shutter press simply resulting in the selection of the most recent sample. But I guess that’s negated by long exposure times.

    One test to perform is to take pictures of non-moving terrestrial subjects under similar conditions. Another test would involve the North Star. Even though it’s not perfectly centered, it would be interesting, nonetheless.

    Also, does that camera have a longer timer interval available to reduce the effect of the supposed tripod oscillation?

    1. >I’m going to guess that the IS is trying to stabilize during the entire time of the exposure
      > and sees the apparent motion of the stars (or Moon) due to the Earth’s rotation as if it
      >were camera motion.

      Strangely enough, it *is* camera motion– the camera is strapped to the earth. All of the motion of the stars that we can see is due to the rotation of the camera (and earth), not the stars moving around with respect to each other and us.

      >Atmospheric interference may also come into play.

      Not significantly– the image with IS turned off has very straight lines after the camera stops wobbling, so atmospheric distortion is apparently negligible.

      >It seems hard to believe that the IS is this “smart” though. I had always assumed that IS
      > in digital cameras was a simple matter of periodic sampling as if the shutter were
      >snapped at regular intervals with the actual shutter press simply resulting in the
      >selection of the most recent sample. But I guess that’s negated by long exposure times.

      The correction is definitely active in the sense that it uses a moving lens to correct for motion of the camera. The real question is whether this system works on a tripod or not– and what is it actually doing? Is it detecting motion of the camera or of the image? Whatever it’s doing, it *is doing something* and not what I would expect. The wikipedia article suggests that the camera should be smart enough to detect that it’s on a tripod… but that’s not obvious either. (I see no evidence online that digital image processing is used to stabilize the image, and I find that surprising– the problem is *identical* to that of working an optical mouse.)

      >Also, does that camera have a longer timer interval available to reduce the effect
      >of the supposed tripod oscillation?

      I think that I was using a 4 second delay; longer is indeed possible.


      Windell H. Oskay
      drwho(at)evilmadscientist.com
      http://www.evilmadscientist.com/

      1. >The correction is definitely active in the sense that it uses a moving lens to correct
        >for motion of the camera. The real question is whether this system works on a
        >tripod or not– and what is it actually doing? Is it detecting motion of the camera or
        >of the image? Whatever it’s doing, it *is doing something* and not what I would
        >expect. The wikipedia article suggests that the camera should be smart enough to
        >detect that it’s on a tripod… but that’s not obvious either. (I see no evidence
        >online that digital image processing is used to stabilize the image, and I find that
        >surprising– the problem is *identical* to that of working an optical mouse.)

        To see what I mean about digital image stabilization, turn on IS and pan the camera without pressing the shutter button while watching the screen. On most of the digital cameras I’ve played with that have IS, the image jerks from frame to frame. I suppose optical IS could look like this, but it just seems more like a digital effect to me.

        According to the article, optical IS detects motion of the lens while it seems to say that digital IS detects image motion.

  2. First of all, I wonder if the IS can tell (and if so, whether it uses) which items are in the far field. Those parts of the picture are more likely to be stable, so it probably tries to stablize wrt to that.

    Second of all, the Earth’s rotation is not only minor over a few seconds, but perfectly linear. Star twinkle from the atmosphere, however, is fairly large and randomish. If the IS is trying to get all the star images aligned at the same time but they are (apparently) moving around independently, you are going to get a squiggle.

    Here’s a test: If it’s inaccurate tracking, you should get pretty much the same squiggle in every picture (of the same length). If it’s jitter caused by twinkle, you should get different squiggles in each picture.

    Another test would be images of different lengths. If the problem is the tracking, a longer image will result in a larger diameter squiggle. If the problem is twinkle, the squiggle size should be pretty constant.

  3. I don’t know if it’s of use or interest, but I managed to get OK pictures of the lunar eclipse on 03/03/07 with a Panasonic DMC-FZ20 by

    a) using a tripod
    b) Turning image stabilisation off
    c) Using the 10 second timer
    d) setting the exposure to 1s
    e) setting the aperture to 2.8

    I live in London, right by a very well lit road (road light beams in through my bedroom window!), and the light pollution is horrible. The sky is perpetually slightly orange.

    There are more details, and the photo is here: http://www.flickr.com/photo_exif.gne?id=410280095

    It definitely could be crisper – it is only very slightly less blurry than yours, but I think a little more detail is visible.

  4. Haha, you guys are crazy.

    Assuming your IS works the same as mine [Canon 28-135 EF IS] it is being ‘stabilized’ by a gyroscopic lens element. The gyroscopic element adjusts to compensate for hand shake and does quite well. However, if you’re using a tripod (which you have to be, unless you have the steadiest hands I’ve ever seen — a one second exposure?) the gyroscope has nothing to compensate against and therefore spins and vibrates uselessly, compounding blur. If you were hand-shooting I would be very, very impressed if you got sharper images without IS than with it.

    1. Wow — I suppose I’m wrong. I read your update. I guess you’re shooting with a point-and-shoot rather than a dSLR? Hummm.

      It looks like other methods of non-gyroscopic IS try to adjust the image itself frame-by-frame to match the input. Glad my camera doesn’t work like that. Although, you would think that’d be <i>most</i> effective in a night sky with points of light.

  5. So, instead of speculating, I went looking…

    According to Canon and Steve’s Digicams the S3-IS uses an optical system similar to Canon’s EF lenses. This confirms what Windell said.

    Is it possible that the IS is sensitive to the acceleration due to gravity or to Centripetal Acceleration due to the Earth’s rotation? If so, this would seem to be undesirable under any circumstances.

    to try: Does the shape of the squiggles change if the camera is pointed in different directions relative to the rotation of the Earth? It might be particularly interesting to take a long exposure of the North Star.

    In the advanced manual for this camera, on page 40, it says “Camera shake may not be fully corrected when shooting with slow shutter speeds…a tripod is recommended.” But that’s what you did.

    On page 57, “Please note that camera shake becomes a factor at low shutter speeds and with larger aperture values…Use…image stabilizer…Raise the ISO speed…Raise the flash…Use a tripod”

    to try: Try setting the IS mode to the other settings (page 39) to see if the results are different. (available modes: continuous, shoot only, panning)

    [the comment software doesn’t seem to like links with equal signs or ampersands in URLs in the link: tag, so some of the links in this comment may not work so I include them below as plain text]

    Canon: http://www.usa.canon.com/consumer/controller?act=ModelFeaturesTabAct&fcategoryid=223&modelid=13077&pageno=1

    Windell said: http://www.evilmadscientist.com/comment.php?mode=view&cid=2390

    advanced manual: http://alpha02u.c-wss.com/inc/ApplServlet?SV=WWUCA900

    1. Okay cool– things to try. (* It may take me a while to get around to it, however!)

      It looks like the canon advanced manual cannot be directly linked to; to get to it, you need to browse through the site to find it in the list of literature related to that camera.


      Windell H. Oskay
      drwho(at)evilmadscientist.com
      http://www.evilmadscientist.com/

  6. Hmmm…I’ve been wondering about this very thing. I recently took some night pictures with my Canon S3 while vacationing at the beach. I had the EXACT same strange, squiggled blur on some of my street pictures that you have in your star pictures. I used a long exposure and the camera was not on a tripod, but sitting on a balcony ledge. The blurs were on light emitting from stationary objects like buildings, street lamps, etc. I thought something was wrong with my camera at first, or that the wind was shaking it, but here I see that you’ve had the same problem with the stars while using a tripod. Did you ever figure out what causes this, if it indeed was the IS, or the gyroscopic lens element? Do you think there is a solution, or will this model camera always do this when taking night pictures? Thanks for this post, good to know it’s not just my inferior photography skills!

  7. There seems to be a general consenus among pro/amateur photographers that Image Stabilization should be turned off when using a tripod, as the gyroscopes overcompensate for any real/perceived vibration and create a "feedback loop" whereby they begin to compensate for their own adjustments.

    This idea is further cemented by the fact that the latest generation of Canon "IS" lenses have built-in sensors that detect whether the lens is mounted on a tripod, and if so, deactivate the IS feature.

  8. Maybe this can be helpful for you:

    At this web address: http://picasaweb.google.com/mriscoc/MisFavoritas/photo?pli=1#5084339379296164562 you can see a photo of the moon taken without tripod, using a Canon S2 IS, maximum zoom (included Digital). Because the maximum zoom? for get an appropriate value of exposition, you can see the exif value of this picture:

    ISO:400
    Exposición: 1/125 sec
    Apertura: f/4.0
    Longitud focal: 72mm

    Note the exposition time: 1/125 s not was necessary a large time. To take a eclipse photo maybe is necessary more time because lack of light, maybe a 1/30 would be enough

    Miguel A. Risco
    http://mrisco.accesus.com

  9. Image stabilisation ( and not only an image) is a quite an easy proccess.
    Your camera have an acceleration detector. It can give your camera information about it’s movements in some range of speeding-up.
    Camera process that information and convert it to stabilizer movement.
    If camera moves down, for instance, your stabilizer moves up, compansating that parasitical movement.

    And there it one hidden thing – the quality and capabilities of components.
    The detectors could be accelerometers or gyroscopes.To make precise photos Habble has excellent quality gyroscopes.
    Your camera of course don’t has. It has something similar to accelerometers .
    And the accelerometers could not catch slow movements, it has not enougth sensetivity.

    So, if you take a long exposure shot with tripos (or same solid base), camera can’t get enought precise information about it’s position. It gets low signal + noise from accelerometers.
    Therefore, the strange IIS compensation on the picture is the result of IIS noise processing.
    Longer exposure gives summation of more noise processing errors amount and worse quality.
    In usual case, with short exposure it works perfectly.

    PS sorry for my English

Comments are closed.