[Gimbal]

Really? I would be surprised if this construction is done with anything else than leaf springs (which would mean frictionless movement), but then again, I never saw a suspension system for a camera sensor. Wouldn't it just be that you don't hear it moving when it moves slowly?

At least the K10 (and probably every other Pentax as well) has the sensor plate pressed between a sort of ball bearings. With this method they make sure that the sensor does not move in any other direction then the up/down left/right plane. But it also means that there is mechanical friction involved.

And when there is mechanical friction and a simple PD regulator (but that is, as I understand it, only speculation from Lumolabs?) involved you will have problem with starting friction when there is a small error between wanted and actual position. Since the error signal in such case is small the correcting force will also be small and might not overcome starting friction.

[falconeye]

1. I wonder if there's a human factor involved too, not just the camera? The blur seems to me to be a harmonic disturbance. Harmonics are generated under certain conditions. Change the conditions and the results change. The fact that a good tripod eliminates the problem points to the fact that grip is significant, and so is the 50% margin measured in the test.

2. I am an excellent shot with a rifle. I believe some of that technique also relates to photography. I tested my K-x at 1/60, 1/125 and 1/250s and none exhibited blur. That may be because my camera is different somehow, or maybe I hold the camera differently (softer, harder, looser, steadier?). Or maybe my testing wasn't thorough enough.

3. I suspect someone will use this work to criticize in-body shake reduction.... Fast shutter speed generates oscillation in the sensor, causing blur in the image.

Ad 1. We managed to specify a testing situation exactly reproducing the effect but eliminating any human factor.

Grip won't help except if you have a veteran iron hand. The shutter moves the body by only ~10µm. The human skin is too soft and too thick to do anything against, whatever be your grip.

Ad 2. We made no statement about the K-x. And yes, thorough testing was what was missing prior to our work.

Ad 3. Yes, but we express the opposite opinion wherever we can. In-body shake reduction is great actually because theoretically it could counteract any shutter blur effect. The K-7 probably can't because the required frequencies are 10x higher. But it could be done in theory so the in-body SR approach is superior.

P.P.S.: Pentax needs to start selling "Class A" tripods. Surely people will buy more of these now. Looking at my username, perhaps I should start selling them?

A 120mm x 60mm x 50mm thick iron plate with a tripod screw and a tripod hole upgrades any tripod to a class A tripod (it should reduce blur width from max. 11µm to max. 3µm and make it neglegible). It will help in muscle building too when shooting free hand

Selling them is a good idea. Shipping cost may be a little high though

Falk, you referred to a "static offset" blur component. If I interpret your charts correctly, it looks to cause on the order of 1px blur. Any thoughts as to it's origin?

Also, did you make any measurements with the camera in portrait orientation? One would expect the extraneous blur component to be midway between the 0 degree and 180 degree results, no?

You need to read the paper on Image Sharpness. In a nutshell: The static blur offset expresses the fact that even a perfect image has blurred edges and you want them blurred to avoid the stair case effect on tilted lines. Moreover, no image is perfect in practice and there will always be significant blur in an image unrelated to shake.

Yes, we assume portrait orientation to be slightly worse than landscape but didn't measure it.

I don't see how this changes the observation that SR was ineffective above 1/100s? Unless your data was simply extrapolated from 1/100s.

It does change the observation. The data wasn't simply extrapolated but at higher shutter speeds, shake became too small to be measurable exactly enough to differentiate. Except for very long focal lengths. And this is exactly how this study was started: by doing a 300mm measurement, first by Rüdiger, then by myself. So, we found it isn't SR but something else and continued to hunt it down.

Next will be to update the SR guide. It is too pessimistic for shutter speeds < 1/200s and long focal lengths.

At least the K10 (and probably every other Pentax as well) has the sensor plate pressed between a sort of ball bearings. With this method they make sure that the sensor does not move in any other direction then the up/down left/right plane. But it also means that there is mechanical friction involved.

a simple PD regulator (but that is, as I understand it, only speculation from Lumolabs?)

The PD controller is our speculation. We had to assume some implementation to make quantitative statements if it could be involved in all of this. A variant of a PD controller was deemed to be the most likely implementation.

Friction may or may not be involved. But note that this alone wouldn't explain the difference between K20D and K-7.

[VaughnA]

Very interesting paper and thanks for the hard work in testing and documenting your findings. Would a grip or other extra weight on the camera change the resonant frequency enough to make a difference? Could the extra weight of the K20 be part of the difference? If so, then the Kx would be even worse.

[opiedog]

seems like an iron man suit would mitigate this issue

seriously, besides a sturdy tripod is there nothing else one can do to lessen the effects?

i noticed that using a flash where flash is required seems to mitigate the issue somewhat.

---

Last edited by opiedog; 07-22-2010 at 04:54 AM.

[falconeye]

Would a grip or other extra weight on the camera change the resonant frequency enough to make a difference?

We don't assume a resonance scenario. There is no periodic excitation. Just possibly a swinging scenario where amplitude is proportional to perturbation magnitude. And the perturbation is the body movement.

So, a grip (preferrably filled with six eneloops) would reduce the effect (by about 30%). It isn't that large anyway so that may already do it.

The K20D is only 4% heavier than the K-7 (with a small lens attached). It is not that the K-7 isn't heavy enough or that small cameras have a problem a priori!

i noticed that using a flash where flash is required seems to mitigate the issue somewhat.

A flash is fast enough to eliminate the blur. If it is the dominant light source.

[mattdm]

The exact amount of average blur is shown in the opening figure of this article. It has its maximum for shutter speeds of about 1/100s to 1/80s. It is less than 5 µm for 1/25s and slower. Or 1/250s and faster.
[...]

In these cases, we highly recommend to shoot at 1/25s (or slower) and to enable shake reduction as it is highly efficient at such exposure speeds. The images will be sharper than at 1/100s!

[...]

Thanks. This is interesting research which is helpful for getting the best out of the camera. It's unfortunate that the firmware isn't end-user tweakable -- if it were open-source, it would be trivial to make the program modes avoid that shutter speed range if possible (which, unfortunately, they seem to prefer).

[Eruditass]

It does change the observation. The data wasn't simply extrapolated but at higher shutter speeds, shake became too small to be measurable exactly enough to differentiate. Except for very long focal lengths. And this is exactly how this study was started: by doing a 300mm measurement, first by Rüdiger, then by myself. So, we found it isn't SR but something else and continued to hunt it down.

Next will be to update the SR guide. It is too pessimistic for shutter speeds < 1/200s and long focal lengths.

If shutter blur is too fast for SR to make a difference, how does it affect the observation at 300mm? Blur would still be dominated by shutter blur (at large magnifications like 300mm), with SR having minimal effect on blur (unless you are somehow separating shake blur from shutter blur, which seems empirically impossible).

[falconeye]

If shutter blur is too fast for SR to make a difference, how does it affect the observation at 300mm? Blur would still be dominated by shutter blur (at large magnifications like 300mm), with SR having minimal effect on blur (unless you are somehow separating shake blur from shutter blur, which seems empirically impossible).

Blur at 300mm and 1/100s is not dominated by shutter blur.


Look, to see this in detail: shutter blur depends on the exposure time and maxes at about 1/100s. But it is almost independent of focal length and of shake reduction in operation or not. Call this blur B(t).

Free hand shake is mostly from camera rotation. Mostly proportional to both, focal length f and exposure time t. It is reduced by shake reduction where SR may reduce this to say 1/3. Call this blur A. With A = a f t. And shake reduction may reduce a to a' = a/3 or so.

Together, you have B(t) + a f t or B(t) + a' f t.

So, if you measure the sum, it may be that B(t) is too large to see the difference between a and a'. And because you cannot increase t (B(t) becomes 0 for large t) you have to increase f to see it. This is what you do when using a 300mm lens.

[Winder]

To throw something out from a previous experience. In a previous life I encountered a problem where a small electric motor was inducing a slight harmonic vibration into an aluminum chassis and causing some annoying problems. When bolted down the vibration was eliminated (as expected). Since the unit was moved to different locations frequently we had a machinist make our version of a quick release plate. We noticed that the when just the quick release plate was attached the harmonic vibration was canceled.

Finally we noticed that with just the quick release screws installed (no plate) the vibration was canceled.

The screws where threaded up into the chassis of the unit and simply by tightening the screws into the chassis we stressed the chassis enough to remove the vibration.

I'm not sure if this would have any effect on a K-7 of if the threads for the tripod mount are machined into the chassis.

[traderdrew]

I know your study says it only effects wide angle lenses. I'm not trying to argue with you since you put a lot of work into it.

I used a DA*300mm lens during a cloudy morning to shoot a butterfly at 1/100th of a second with no tripod. The image is on facebook -

Fan photos from PENTAX | Facebook

This is a crop of the same image:

Fan photos from PENTAX | Facebook

[falconeye]

I know your study says it only effects wide angle lenses. I'm not trying to argue with you since you put a lot of work into it.

I used a DA*300mm lens during a cloudy morning to shoot a butterfly at 1/100th of a second with no tripod. The image is on facebook

Actually, the study says that it is most easily noticed with wide angle lenses.

Your 300mm image is spectacular. Nowhere in our study do we say that such images aren't possible. Among several images, the shutter blur can be smaller than 5 µm even at 1/100s (and it affects horizontal lines) and you can be on the lucky side with free-hand shake and SR perfomance which did have a chance to reduce free-hand shake in this situation.

So, you did the right thing and took a number of captures in this delicate situation.

[bertbert]

Great work! Let me also state that I enjoy your blog, it's a very good read!

[nanok]

wow. this is not only very respectable and useful work, it is also a hell of a read on it's own (i have yet ot read the full article though, a bit busy here); this comes as no surprise, for people who have seen to way you approach things/argue before. what a treat .

anyway, i wanted to ask something: as this is (relatively) close to the sync speed, could this have something to do with the stopping action of the first curtain (and the very near, in time, starting action of the second curtain)? strangely enough, the sync speed didn't change, but maybe something else in the mechanics of the shutter changed, making it more "brutal" when it comes to that final/starting deceleration/acceleration. it sure sounds like it, apparently other shutter speeds have the two critical points (stop and start respectively) too far apart to contribute to a major effect (slower ones will have the second start delayed, faster speeds will have the starts increasingly closer together -- traveling slit -- , so further between first stop and last start as you go "faster"). does it make any sense?

[dlacouture]

...(unless you are somehow separating shake blur from shutter blur, which seems empirically impossible).

I'd say that shake blur is slow and of great amplitude, whereas this shutter blur is ultra fast and small, so they are pretty different.

Another thing to take into account is that this effect will be smaller for every addition of weight, and long lenses often are heavier than wide angles.
I've seen this blur from time to time with my 18-55 (especially in the portrait orientation), but never with the hefty 70-200/2.8...

Furthermore, from memory, the SR sensors can truly compensate for "rotational" shake only, not translations (I don't even know if gyros are capable of accurately detecting a translation!)...

[nanok]

i think they are using accelerometres to be precise (not gyros, there would be no room really ), and they do provide information to detect translation. this is how they work actually, by detecting + predicting + correcting translational movements, through compensating opposite translational movements of the sensor