camera shake - quantifying?
bugbear
and "depth of field", and quite a bit on "diffraction limits".
(e.g.
http://www.wrotniak.net/photo/tech/dof.html
)
Can anyone point me at a reasonable technical
discussion on camera shake in relation
to pixel errors?
In particular resolution versus shutter speed?
Clearly (!?) a very low res camera (.e.g 640x480)
is much more tolerant of camera shake than a Megavision
(
http://www.findatlantis.com/wiki/index.php/Keeping_it_Sharp
)
Since camera shake can be reasonably expressed
as angular movement of the camera, the relationship
between pixel count and lens length appears key,
but the maths is beyond me.
I know the old rule says 1/focal length as shutter speed
for hand held, but that rule (it appears to me) ASSUMES
some constant sensor resolution.
Can anyone point me at an analysis?
BugBear
Paul Furman
Print size is part of the calculations so the DOF calculators including
that are the place to start. Low-res sensors mean smaller print sizes. I
don't know what number to assign to camera shake, it's different for
each person.
--
Paul Furman
www.edgehill.net
www.baynatives.com
all google groups messages filtered due to spam
ArlanHansen
wrote:
I can't (because it's never been a concern of mine, so I never researched it).
But I can tell you that sub-pixel-width shake can be used to great advantage to
create ultra-high resolution photos, far surpassing the camera's available
resolution. When using programs like PhotoAcute Studio you are able to use
stacked images that vary in sub-pixel differences to greatly increase all detail
in any scene. For a quick sample see:
http://www.photoacute.com/studio/examples/screenshot/index.html
I thought it was worth mentioning in your quest for camera-shake info. You'd be
surprised what kind of lemonade you can make out of previously-thought-of lemons
with a little knowledge. :)
This is why I never throw away any of my photographic "accidents" anymore. I
stopped hitting the occasional "delete" when reviewing my photos for the last 5
years. Between utilities like this and Fourier-transform utilities to refocus
blurry images or undo camera-shake on single frames, who knows how those "fatal
photographs" can be put to great use 10 years from now. The bokeh from stacked
images taken from only one vantage point can even be used to create 3D scenes
when using simple utilities like Picolay freeware. Ain't digital wonderful?
Chris Malcolm
> I know the old rule says 1/focal length as shutter speed
> for hand held, but that rule (it appears to me) ASSUMES
> some constant sensor resolution.
> Can anyone point me at an analysis?
It also assumes a certain value of hand shake, which varies from grip
to grip and person to person, and possibly with temperature and
certainly with wind. I personally found it a useful rule for images up
to about 3MP and A5 print size, which I had to adapt proportionally
for higher resolutions. Your camera in your hands is very likely
different. You don't want analysis, you simply need to experiment with
your own hands and camera.
--
Chris Malcolm
bugbear
Wondefully interesting, and delightfully irrelevant to my question.
Splendid!
BugBear
bugbear
Actually I do want analysis, since I want to be able
to make predictions about equipment to buy that I don't already
own.
And I'm not just talking about hand-held.
With long lenses and high res sensors,
one needs to consider HOW rigid a tripod
is needed, since most tripods are not absolutely
rigid, merely rigid enough for typical use.
BugBear
Don Stauffer
results of camera angular vibration. It still takes a lot of math to use
it- one must multiply the rest of the system mtf by the mtf of image
motion, for each spatial frequency in the range.
This curve was done by Perkin Elmer, who makes reconnaiscence cameras
for the military.
Don Stauffer
Most image motion analysis I have seen simplifies and assumes a
sinusoidal motion of some given amplitude. That was originally done for
vehicle vibration in camera sensors mounted to airplanes, but I suspect
that a sinusoidal vibration is not too bad an approximation for hand
vibration also. The key again is the amplitude estimate.
Chris Malcolm
The flimsiest tripod is perfect if standing on a completely stationary
surface and the shutter is activated remotely. How wobbly a good
tripod is depends on the forces applied and their timings. With a long
lens and a high res sensor no tripod will fail to wobble and blur the
image in a high wind. And how much it will wobble will depend on the
direction of the wind, where the mount is with respect to the effective
centre of force given the wind resistance characteristics of that lens
on that camera at that direction of wind, etc. etc..
I can't imagine a formal mathematical analysis which would be any use
here, given how many important parameters would be unknown in
practice.
--
Chris Malcolm
paul tander
You must also take into account the type of camera used. All DSLRs will, by very
design, have to shake the camera and tripod during each exposure. This is due to
the inertia and impact of mirror and focal-plane shutter. Whereas a P&S camera
generally uses a silent electronic shutter and balanced leaf-shutter system.
Imparting virtually no extra camera shake on each exposure or so little that
it's practically immeasurable. Some DSLRs allow you to lock up the mirror to
prevent some of this excessive shake that their body design causes, but you
still have to deal with the focal-plane shutter's start-stop shake. The downside
to using mirror-lock-up: unless your DSLR also has live-view you won't be able
to use the OVF for composition or watching when to trip the shutter at the right
time while the mirror is locked out of the way.
For absolute minimum camera shake out of all possible scenarios it is best to
use a high-quality P&S camera and either a remote-shutter release (see CHDK's
inexpensive USB-Remote system for example) or use its built-in self-timer to
trip the shutter. Due to the smaller mass of most P&S cameras you may also use a
much lighter tripod for maximum stability than you would need with any DSLR.
Using any DSLR to minimize shake is out of the question. That won't ever happen,
except if used in only one manner ...
If using a DSLR, you *may* obtain shake-free images by using an old method once
popular in astrophotography circles, used by those capturing images with an SLR
attached to their telescope. However, it is only useful for long exposures of at
*least* 1 second or more. It's called the "hat-trick" technique and was intended
to be used over the end of the telescope. Cover the lens with a hood (a deep
felt hat that the astronomer wore was commonly used long-ago, hence its name).
Do not to touch the camera with the hat/hood, but block out all light from
entering the lens. Open the shutter. Wait about 10 seconds to allow all
vibrations to die down. When using long telephoto lenses oscillating vibrations
can be seen up to 30 seconds or more depending on tripod design and stability.
In higher powered telescope setups then oscillations can be seen for up to 1
minute after tripping an SLR shutter. Remove the "hat" covering. Count-off your
exposure. Replace the covering being careful to not touch the camera until the
lens is adequately covered. Close the shutter. Sounds absolutely prehistoric
doesn't it? Well, that's what you have to do when shooting with those
prehistoric focal-plane shutters and flapping mirrors if you want to prevent all
camera shake from ruining your image resolution, even in today's DSLRs.
Alan Browne
> Since camera shake can be reasonably expressed as angular movement of
> the camera, the relationship between pixel count and lens length
> appears key, but the maths is beyond me.
It's complex movement (several orders) coupled to the jerking motion
when the shutter is depressed (why a gentle squeeze is needed).
> I know the old rule says 1/focal length as shutter speed for hand
> held, but that rule (it appears to me) ASSUMES some constant sensor
> resolution.
The 'rule of thumb' speed supposes prints of about 5x7 at largest.
Maybe 8x10 if you're really calm.
> Can anyone point me at an analysis?
I guess one could equip a camera with accelerometers and rate sensor and
collect a model and then publish some empirical formulae to simulate the
movement. I don't know of any such. Perhaps the camera makers have
this but don't publish it.
If you look at dpreviews recent a900 test they explain their methodology.
Here is my test of the Maxxum 7D from a couple years ago:
http://www.aliasimages.com/KM7D_AS_Test.htm
Should be repeating that for the a900 sometime in November.
--
-- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
-- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
-- [SI] gallery & rulz: http://www.pbase.com/shootin
-- e-meil: Remove FreeLunch.
-- usenet posts from gmail.com and googlemail.com are filtered out.
Dave Martindale
>Can anyone point me at a reasonable technical
>discussion on camera shake in relation
>to pixel errors?
The math is pretty simple *if* you have measurements or a reasonable
assumption about the amplitude of the shake.
To convert an angular rotation of the camera body into the length of
line on the film plane it will be blurred into, just remember that the
"chief ray" of a bundle of light rays goes into the center of the front
of the lens (really, the front principal plane) and comes out the back
center of the lens (the real principal plane) *without changing
direction*. So one degree of movement of the camera around its axis
causes the ray from the rear of the lens to change direction relative to
the camera body by one degree too. Also, for subjects near infinity
(really, anything more than a few focal lengths away), the rear
principal plane of the lens is almost exactly one focal length away from
the sensor.
For example, continue to assume 1 degree of rotation. Also assume that
the lens focal length (the true focal length, not the "35mm equivalent
focal length") is 50 mm, so the sensor is 50 mm from the lens rear
principal plane. Then, for an object near the centre of the image,
rotating the camera 1 degree causes a displacement of 50 * tan(1 deg) =
0.87 mm. (The correct formula is actually slightly more complex, but
for small angles this one is equally good).
That's in mm. If you want an answer in pixels, take the length of the
displacement and divide by the pixel size. For example, 0.87 mm is 870
um, while a DSLR sensor might have 6 um pixel pitch, so this length is
145 pixels.
>In particular resolution versus shutter speed?
To take shutter speed into account, figure out the maximum angular
rotation rate for a shake, then multiply by the shutter period to
determine the worst-case angular change while the shutter is open. Then
apply calculations above.
If you don't know the maximum rotation rate, but you do know the peak
amplitude of the shake and its frequency, assume the motion is a sine
wave and apply a little calculus to get the maximum rotation rate.
>Clearly (!?) a very low res camera (.e.g 640x480)
>is much more tolerant of camera shake than a Megavision
Yes, because the inherent unsharpness of the former camera's images
means you will print them small, and it will take a larger amount of
blur to become visibly less sharp.
>Since camera shake can be reasonably expressed
>as angular movement of the camera, the relationship
>between pixel count and lens length appears key,
>but the maths is beyond me.
See above. The critical things are focal length, shake rotational
velocity, and shutter speed.
>I know the old rule says 1/focal length as shutter speed
>for hand held, but that rule (it appears to me) ASSUMES
>some constant sensor resolution.
It assumes a particular size camera, in this case I think 35 mm.
Now, it's true that no matter what size the camera and film, for a
particular amount of shake, using a shutter speed of 1/focal length
gives you the same length blur in the image plane. But the same length
blur is a smaller proportion of a larger format sensor or negative, and
if that negative is enlarged less to produce the final print, the rule
can be relaxed a bit. Also, larger cameras tend to be more massive and
shake at lower frequencies, and for a constant shake amplitude, the peak
velocity is proportional to frequency. So larger and smaller cameras
are likely to need a slightly different constant in the rule of thumb.
Dave