Re: Averaging preprocessor

> There are more principled approaches to this ... a popular one is to
> perform conservative spatial noise filtering followed by ME and kallman
> filtering of pixels along the motion path. This has been done by a lot =
of
> people, one example is [...]

Ah, interesting paper indeed.. It's certainly much better an approach for=
=20
moving areas than averaging, but for static ones, I'm still not so sure. =
If I=20
understood correctly, the proposal in the paper minimized number of back=20
buffer frames by refining a statistical noise model for each frame. The=20
drawbacks are that while the scheme works almost as well for moving block=
s as=20
it does for static ones, the noise is assumed to be gaussian white and th=
at=20
using a spatial (even though paramerized) filter causes blurring.

The big plus in averaging static blocks is that it would effectively redu=
ce=20
*any* kind of noise, including DCT artifacts when recompressing stuff. It=
=20
won't help the blurring, though, if the camera shakes a little bit.

If you had to choose a one, the Wiener/Kalman stuff would probably be bes=
t=20
general purpose filter, but a combination with averaged static frames mig=
ht=20
be worth trying, too.

- Jarno

BTW. Required back buffering could perhaps be minimized by allocating=20
placeholders for the averaged result and halting the averaging process of=
 a=20
block once someone decides to start moving it's contents. On the other ha=
nd,=20
it would involve interfacing with motion compensator and might cause the=20
number of frames involved in averages to be too small in general.

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to '[email protected]'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.