Super Steady Shot with Full Frame Sensors
February 1st, 2008 - 12:00:00 AM:
In Internet forums you often hear the claim “Sensor based image stabilisation (Super Steady Shot) does not work with full-frame sensors”. The two main arguments are:
- The mirror box is too small. The sensor has no room to move.
- The image circle of existing lenses is too small. The sensor would move outside the image circle.
The first argument is complete nonsense. SSS is not added to an existing camera. The camera is a new development. There's no law that forbids to make the mirror box a bit wider and higher.
The second argument requires closer analysis. All depends on two factors:
- How far does the sensor have to move?
- How large is the image circle really?
To answer the first question, you can devise a simple test: Use a camera without image stabilisation (or turn it off) and make a long handheld exposure. Then check how far the image has moved across the sensor during exposure.
I've used the Sony A700 with SSS turned off, the 100/2.8 Macro, focused at intermediate distance, and made a handheld 1 second exposure. Note that 1 second at 100 mm is far beyond the effectiveness of 2 to 4 stops that most makers claim. With a 2 stop gain you could shoot at 1/25 second, with a 4 stop gain you could shoot at 1/6 second. For 1 second you'd need between 6 and 7 stops of gain, which currently no maker can provide.
|Test with 1 second exposure, 100 mm|
But still, when checking the resulting image, it can be seen that the image only moved ±30 pixels during the shot. This translates to an absolute movement of about ±0.17 mm. In other words, the sensor would have to move ±0.17 mm to counter the image movement.
Now what about full frame? You can do all the math to translate the image movement to angular movement, then relate that to the angle of view of lens and APS-C sensor, compare with the angle of view of the same lens and a full frame sensor, translate back to the image movement on the larger sensor, and - surprise, surprise - you'd end up with the very same value. This is easily understandable. It's the lens that creates the image, and the same angular movement of the same lens causes the same shift of the image on the sensor, no matter how large the sensor is.
Now let's be generous. Let's say we're using a 600 mm lens. It has a 6 times narrower angle of view, which causes a 6 times larger image shift for the same angular movement. With our 1 second exposure the sensor would have to move ±1.02 mm to compensate. Remember that we're operating way beyond the range of current systems. But we're extra generous and assume that the photographer had two espresso with his lunch, so he'll shake a bit more, let's say ±1.5 mm. Now take a step back. Does anybody really think it's impossible to move a 36×24 mm sensor ±1.5 mm for image stabilisation? And does anybody really think you can't make the mirror box 1.5 mm larger in each dimension to give the sensor the room to move?
Now let's look at the image circle of existing lenses. Most people who claim that it's not large enough appear to never have seen it. With all lenses the image circle is larger than 43.3 mm in diameter. The 36×24 mm frame is not that tightly squeezed into the image circle. Furthermore, the edge of the image circle is not a hard cut-off. It's a rather soft transition. Especially with telephoto lenses, where you need the largest sensor movements because of the narrow angle of view, there is quite a wide image circle with little vignetting. Vignetting is generally a larger problem with wide angle lenses, but with these you also need only small corrective sensor movements.
Also consider lenses with a smaller image circle designed for APS-C cameras. With these SSS works just as well as with full frame lenses.
The sensor has to make only small movements to counter shake, and it's
independent from the sensor size. Furthermore, the image circle of existing
lenses is large enough for sensor-based image stabilisation, even for full-frame
© 2008 Michael Hohner
|#1: Comment posted by Paul Cassidy on March 3rd, 2008 - 04:18:37 PM:|
After reading through this it certainly seems possible that a full frame is coming soon and with the SSS feature intact. Now if they would only go to 100% full frame viewfinder? I hate finding surprises on the edges of my images when I get home. I'm a very old Nikon F user and have always had the 100% finder ;-)
Also wanted to take this opportunity to say THANK YOU! for such a wonderful web page with so much great data about my Minolta Maxxum 7 and SONY Alpha100 systems. I'm new to the a-mount world and find myself visiting here more and more as my system grows and question come up.
|Michael Hohner answers:|
|Sony has already announced that their Flagship camera will be full frame and will have SSS. They have not said anything about viewfinder coverage, though.|
|#2: Comment posted by Ardana W on May 29th, 2008 - 10:26:38 AM:|
|Dear Herr Hohner,
First of all, thank you and congratulation for having such a great site with such attention to details. I found it certainly VERY helpful.
Could you please explain to me as to how you would come up with the value of ca. 0.17mm for a movement of 30 Pixels for the sensor? Did you multiplied it with the size of the individual Pixels? and if so, where could one find such detailed technical specification on Minolta/SONY cameras?
|Michael Hohner answers:|
|The A700 has 4272 horizontal output pixels, and the horizontal size of the sensor is 23.5 mm. This means that 30 pixels are 30/4272*23.5 mm = 0.17 mm (rounded).|
|#3: Comment posted by olivier Hericord on June 3rd, 2008 - 09:15:43 PM:|
|the problem is perhaps more the inertia of the sensor and it's holding structure (with connections) that has to move quickly and precisely.
in an other hand it's also difficult to make a fast stabilized lens because with a fast lens there a lot of big glass pieces to move.
But SSS is HYPE....SSS can't stop subject movement ....nothing can replace a fast lens
Increasing sensitivity of sensors coupled with fast lenses is the Real solution that will allow shutter speed to be reduced to a point where SSS is useless
|Michael Hohner answers:|
|True, SSS can't stop subject movement. But it's far from useless. A fast lens doesn't help you if you stop down for increased depth of field, and high ISO doesn't help you if you want low noise. In many situations there is no or very slow subject movement, and SSS and other forms of image stabilisation enable you to shoot with the ISO rating and the depth of field that you want. Any type of image stabilisation is a big help, which everybody can confirm who has used it.|
|#4: Comment posted by Stuart on July 3rd, 2008 - 11:17:39 PM:|
|Part of photography is about being creative. Without image stabilisation you are limited to using faster shutter speeds only, freezing motion when sometimes you might want to show some motionblur. That restricts the creative process and inhibits some artistic techniques (such as "pushing the shutter").
Part of lowlight photography specifically is about making compromises in order to get a decent exposure. High ISO, flash, and large apertures are possible compromises, which each come with their own limitations and drawbacks. A slow shutter speed is another possible compromise, with it's own limitations and drawbacks. None of them should be dismissed outright. However; without image stabilisation, a slow shutter speed is not possible, so you have limited your available options.
|#5: Comment posted by falconeye on July 6th, 2008 - 01:36:17 AM:|
thanks for the interesting argument. There is an even simpler argument:
No blur with the rule of thumb (t[s]=1/f[mm]) means that the blur has the size of a standard circle of confusion (0.03mm for 35mm full frame). If SSS has to gain 4 stops, it means that the blur (which needs to be compensated) becomes 2^4*0.03mm = 0.5mm. For any focal length.
The big challenge is the time BEFORE you press the shutter while the SSS accomodates. During this time, the sensor is already shifting and may travel by much longer distances. It would introduce an extra latency to first re-center the sensor when you press the shutter release.
|Michael Hohner answers:|
|If you listen to your camera you notice that the sensor does NOT move before you fully press the release. You can clearly hear SSS in action, and it starts when the mirror flips up, not before.|
|#6: Comment posted by Marko Björkroth on July 17th, 2008 - 05:41:53 PM:|
|I agree, image circle is not a problem. The real problem is the shutter port.
According to the specs, the AS mechanism of D7 and D7 can move the sensor up to 11 mm diagonally. Assuming the range of movement is identical in horizontal and vertical direction, the sensor can move up to 4 mm away from the centre location (both horizontally and vertically).
We also know that the size of the active (image recording) area is quite a bit smaller than 35 mm film frame - just 23.5x15.7 mm. However, the shutter port appears to be full size (roughly 36x24 mm) which means there's 4-6 mm clearance around the CCD's active area - exactly as much as the CCD is allowed to move!
Of course, it makes no sense to move the CCD any further because it would get partially under the edge of the shutter port.
But how it's going to work for full frame? I think there are two alternative solutions:
1. Build an oversize shutter mechanism. And, probably, a bigger mirror box, too.
Sounds expensive, but the cost might not be prohibitive if the system is designed for a smaller range of movement (+/-2 or maybe just +/-1 mm instead of +/- 4 mm). As you have pointed out, over 1 mm movement is rarely needed - I think we got this +/- 4 mm range because it was achived without any additional cost, simply by using parts designed for 35 mm film cameras.
2. Use full frame only when AS/SSS is off and 1.1X crop when AS/SSS is on.
|Michael Hohner answers:|
|The mirror box and shutter are not standard parts taken out of a big box. The mirror box is not a part in itself, anyway, and the shutter is designed and built specifically for a camera (or a number of cameras). So adapting these for a camera is not a problem, and especially not for a camera of the price class of the upcoming flagship.|
|#7: Comment posted by Steve Jones on November 8th, 2008 - 02:52:20 PM:|
|A nice rational analysis - I did the same test myself to see how much correction is required and came to about the same sort of result - about 90% of my hand-held shots taken 4 stops slower than the "1/f" rule could be corrected be < 0.3mm.
Incidentally, there is another myth that the absolute movement with a longer lens is greater than for a shorter focal length. It is true that, for a given absolute angular deflection, the sensor has to be moved further with longer focal lengths. However, the stated stabilisation improvements are relative. That is a three stop improvement might allow a 125mm shot to be taken at 1/15th instead of 1/120th, whilst at 500mm it would allow for 1/60th instead of 1/500th.
It we make the reasonable approximation that for small, short, tremor movements the the rate of angular displacement is reasonably constant over the length of typical exposure times we might reasonable want to stabilise, then the amount of physical movement of the sensor is identical in both cases. What changes is the speed at which the sensor has to be moved. Moving the sensor faster, with precision, is a technical challenge, but it's not a theoretical limit such as moving outside of the image circle would be.
Of course actual tremors aren't as simple as that - human and tremor no doubt has lots of random elements, some faster than others, but no stabilisation system copes with everything. With stabilisation we are normally all about improving the "hit rate" rather than a guarantee (the 1/f rule doesn't provide that either), but as a quick simplification I think it's reasonable.
I mention this "myth" that sensors have to be moved further (rather than faster) on longer lenses because no less an authority than Canon issued a white paper which was carefully worded to give the impression that with a 300mm lens, sensor-based stabilisation required a movement of up to 5.5mm (which they, very generously to their argument, approximated to "almost a quarter of an inch").
A little thought experiment reveals this is nonsense - if we require up to 5.5mm sensor movement at 4 stops then that implies at the normal accepted 1/300 we might expect one-16th of that, or about 0.34mm or about 60 photosite pitch lengths on a typical current DSLR.
The slippery bit was the way Canon worded this :-
"Short focal length lenses require smaller sensor deflections; 24 or 28 mm lenses might need only 1 mm or so. Longer lenses necessitate much greater movement; 300 mm lenses would have to move the sensor about 5.5 mm (nearly 1/4”) to achieve the correction Canon gets with its IS system at the same focal length."
Well that's the difference between stabilising the exposure time (relatively short) and the framing time (usually very much longer). Canon and Nikon both claim the advantage of stabilising the viewfinder image, but that's not the issue here. Many people were left with the impression that this degree of sensor movement was necessary. Some of these people, incapable of a little analysis, swallowed this whole - and some still believe it and propagate the myth.
The paper is still on Rob Galbraith's website
I really don't think this was a misunderstanding by Canon. I believe they worded this exactly the way they did to trash an opposing technology through word of mouth. The alternative explanation is that it shows a shocking level of misunderstanding by the authors of the document.