>> > The sinc* kernels are very bad at removing noise. IMHO, the
>> > cubic* kernels are better for this job...
I'm guessing that the cause behind Nicholas' HO is that the cubic kernels
remove a lot of the *lower* frequency content --- the taper of such filters
in frequency space is very gradual compared to the sinc:8, for example.
Some unwanted HF is left behind, and all the LF is attenuated to some
degree.
>> That's what yuvdenoise is for :)
>
>Well, sort of. It removes noise temporally. But it would be nice to
>have a spatial lowpass filter in the toolbox also to reduce noise that
>way. Using a simpler kernel like cubic in y4mscaler does this as a
>somewhat unintended consequence. It would be useful to be able to
>independently specify the filter (kernel) bandwidth and the
...
>It should be an easy change, also. What say you, Matt?
I say "It's already on the TODO list." :)
"o add 'blur' option to scaling engine."
The only complicated part is the code-flow to pass around such an option;
that math pretty much takes care of itself.
It's definitely slated for the next y4mscaler release (whenever that is...).
(Hmm... I wonder if there is something beneficial to the cubic-esque gradual
taper, versus blurring/"noise-reduction" via lowering the cutoff frequency
of an ideally sharp low-pass filter.... Just musing to myself.)
-matt m.
-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn
_______________________________________________
Mjpeg-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/mjpeg-users
