Think Your Camera is Advaced?

On Thu, 8 Jan 2009 21:52:58 +0100, Wolfgang Weisselberg
wrote:

John A wrote:

It would be like taking an ultra high speed video, at several thousand
no doubt noisy frames a second, and combining a selected range of
frames to make one clear image.

So what do you do about read noise?

Good question. What have manufacturers done to get it down to current
levels?

Maybe an all-tube camera is the answer.

John A wrote:
On Thu, 8 Jan 2009 21:52:58 +0100, Wolfgang Weisselberg
John A wrote:

It would be like taking an ultra high speed video, at several thousand
no doubt noisy frames a second, and combining a selected range of
frames to make one clear image.

So what do you do about read noise?

Good question. What have manufacturers done to get it down to current
levels?

Black magic and reading out only once per image.

-Wolfgang

In rec.photo.digital.slr-systems Alan Smithee wrote:
"John A." wrote in message
...

Continuous buffering is something that could be possible, although I can't
think of a real life scenario where it would be that useful if you had a
fast camera anyway. You would be already looking through the viewfinder
at
the subject and anticipating the next moves.

I also can't visualise how it would be possible to choose exposure times
in
post processing that would work with moving subjects. The only way would
be
to capture more information than they already do at the moment at the time
of capture. So that would come down to the basics of sensor
sensitivity/capture capabilities, with low S/N.

If for each shot taken you stored a time segment from the buffer in 3D
(X,Y,Time) rather than the conventional 2D (X,Y with accumulated Time)
you could then time-crop the exposure in post. It would be as if you
took the shot with the chosen exposure time.

With before-and-after buffered capture you could also fix a shot where
you pressed the shutter just a bit too late for current cameras. Sort
of a "Tivo" for still cameras.

With advanced processing you might even be able to achieve image
stabilization in cases where it's the subject shaking, not the camera.

I was visualising shooting something like a tennis player where you can
anticipate their moves in advance. But, thinking about it, I can see the
appeal now especially when shooting something fast that isn't as
predictable, say a wildlife photographer who for example wants to photograph
a Chameleon catching a cricket with it's tongue. The photographer could
then just hit the shutter release at the 'trigger point' and the frames
before and after will also be stored to card.

I still can't see how you could time crop exposure in post though, unless
the camera is on a tripod and it's a static shot. If there's movement, the
frames won't line up. Within reason, software could use scene recognition
and the frames could be aligned, but the capabilities are limited.

The capabilities with commercial software may be limited, but much
more sophisticated stuff which could do the job has been running for
at least ten years in some specialised research labs. It's based on
using the 2D images to develop models of the 3D objects which produced
the 2D images. It's sophisticated enough to create 3D models of
architectural interiors from the video stream of a kid running around
waving a video camera, and sophisticated enough to model the moving
bodies of human dancers and athletes.

In fact now I think of it the CGI people like Pixar and the video game
people are using that kind of stuff now. But you can't run it, at
least not in any reasonable length of time, on your whizzo home PC. It
needs more power. But Moore's Law being what it is, we won't have to
wait long for it be runnable in a home PC.

Where the amounts of movement are small, such as trying to keep a
steady focus in a hand held video, there's a lot of computational
short cuts which can be used which do bring that kind of perspective
adjustment and alignment within reach of your home PC. I think some of
the better panoramic stitching programs are already doing that.

--
Chris Malcolm