Hi -
> From: [EMAIL PROTECTED]
> My statement about obliterating edges is only partly correct. My
That's a relief ;) I've been doing captures and encoding from
a few laserdiscs I have left (that haven't and likely will never
be released on DVD) and haven't seen edge obliteration and if there's
a loss of detail, well - it has been a few too many years since I
got a new set of glasses (can't believe I've gone 15 years or more
without breaking the current set! <g>).
But I do begin to wonder what the commercial outfits can do that
make DVDs from analog sources look so good (but yes, I've seen
a lot of transfers that I/we could have done better than!).
> or sporadic noise). I'm sure it's also a lot faster, since a median
> filter is a simple concept (order the surrounding pixel values and
> take the median) that isn't so simple to code up.
Actually the current 'yuvmedianfilter' is about equally cpu intensive
as 'yuvdenoise' - it'll pretty much eat as much cpu as it can get.
> I was using the filter on high-noise inputs with too low of a
> threshold, and so it resorted to averaging most of the time.
Ah, I've been using a higher threshold that's a lot higher than the
default of 2. Seems better values start around 4. Uh, whatever
you do don't bump up the radius too much because "-r 4" is painfully
slow.
> Previously I had used it after yuvdenoise had removed most of the
> gaussian noise, and that worked out a lot better.
Yep - that works well indeed. I run the data thru a moderate
denoising ("-t 6 -l 2" or perhaps even just "-t 6 -l 1") for decent
material and more aggressive settings ("-t 5 -l 3") for ~VHS style
material. Then it's off to 'yuvmedianfilter'.
Uh, it is a bit sluggish - the pipeline looks something like this:
smil2yuv -i 2 foo.smil |
y4mscaler -O sar=src -O chromass=420_MPEG2 |
y4mshift -n 4 |
yuvdenoise -S 0 -r 24 -t 6 -l 2 -b a,b,c,d |
y4mscaler -O sar=src -O size=704x480 |
yuvmedianfilter -t 4 -T 0 |
mpeg2enc -f 8 -4 2 -2 1 -q 4 -Q 1.0 -o foo.m2v
the first y4mscaler is to convert from 4:1:1 to 4:2:0 in the YUV4MPEG2
format, the second y4mscaler takes the center 704 pixels from the frame
(704x480 is a valid DVD resolution. To yuvdenoise the a,b,c,d turn
to black the borders that vary in size depending on the quality of
the source material and 'y4mshift' (my creation) centers the data
within the frame (varies a lot depending on the original source).
y4mscaler's 4:1:1 -> YUV4MPEG2 conversion is superior and doesn't
take too much cpu time (~5%) and that's why the 'smil2yuv -i2 " and
the first y4mscaler stages are used.
Despite all that I can still get between 4 and 5 frames per second
as the final encoding rate. Dual P4/Xeon systems are a Good Thing :-)
> I may hunt around for a true median filter code, just to see what that
> might do.
I did a quick Google search on medianfilter. Too bad my German
is so limited - seems most of the interesting research requires
more than "restaurant German" to read ;)
Oh, might as well toss ina bit of advice for the Canopus ADVC-100
users that live in the US (where the 7.5IRE setup is used) - turn
switch 2 ON! Blacks finally look more, well, black.
Cheers,
Steven Schultz
-------------------------------------------------------
This SF.net email is sponsored by: Etnus, makers of TotalView, The debugger
for complex code. Debugging C/C++ programs can leave you feeling lost and
disoriented. TotalView can help you find your way. Available on major UNIX
and Linux platforms. Try it free. www.etnus.com
_______________________________________________
Mjpeg-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/mjpeg-users
