--- sean <[EMAIL PROTECTED]> wrote: >I'm also trying to convert a series of old vhs >family tapes. I'm using a borrowed canopus advc >300. I'm about 103 artifacts to go before >perfection. Could you post how you are >converting your tapes? The more specific the >better - i.e. actual command lines.
Enclosed is the bash script that I used recently
to clean up a very difficult video tape. (It's a
video-collage piece, made from stuff recorded off
TV, so I've got a couple layers of videotape noise
to punch through.)
Before I get into that, though, let me cover some
basics. Make sure you are using good audio/video
cables! I swear by Monster cables. My 1-meter
composite-video cable, the most expensive one I
could find at Best Buy that day, cost US$60. Note
that I'm using a composite-video cable, even
though my VCR can put out an S-VHS signal. The
issue here is who does the color separation. If I
play a VHS tape and use an S-VHS cable to carry it
from the VCR, then the VCR is doing color
separation. My Canopus ADVC-300 box has
software-assisted color separation; I'd rather use
that than take a chance on what my VCR is doing.
Sound encoding is pretty straightforward. 00.wav
was generated from 00.dv with "smil2wav 00.dv >
00.wav". wavamp16 is a stupid program I wrote
that lets me amplify audio to the point where it
clips. My theory is, normal analog videotape
audio has its own pops and clicks, and normalizing
with them in mind produces audio that's way too
quiet. So I (arbitrarily) decided that one
clipped sample per minute would not generate
audible artifacts. Another stupid program I
wrote, wavcalcamp16, tells me how much I can
amplify an audio sample so that I get a certain
number of clipped samples. It generated the 1.675
value you see in the script. ffmpeg is the only
open-source program I know of that can generate
Dolby AC3 audio, so I use it.
Now on to the video encoding.
"raw2yuv -i 2" generates raw YUV data from DV
data, but with a 4:1:1 color scheme instead of
4:2:0. In English, that means the color plane is
720x480 instead of 360x240, like it needs to be
for DVD video. The issue is, raw2yuv's method of
downsampling color data is low-quality. The
y4mscaler command immediately after raw2yuv, which
I got from Steven Schultz, does a high-quality
color downscaling. I noticed the difference the
first time I tried it -- colors were MUCH more
vibrant.
The "yuvscaler" command just clips off the bottom
few lines of video, which on a videotape always
seems to consist of blurry noise. No sense
encoding that -- you'll never see it, and because
of its lack of predictability, it's expensive for
MPEG to encode it. Note that I don't clip the
left or right edges; I find y4mdenoise does a good
job of resolving the black bars on the sides, so I
don't need to do anything special to them.
The "yuvcorrect" command is a recent addition, and
although it didn't have the desired effect, I
still think it's worthy. All it does is clip the
pixel values to the range needed to display
properly on a TV screen. I run it before
y4mdenoise so that any artifacts generated by
yuvcorrect's clipping can get smoothed away.
The "y4mdenoise" command (here called
"newdenoise", since I'm usually developing it &
rebuild it often) is the most important part. On
a really clean videotape, like something shot from
a VHS-C camera, you can use "-z 1 -t 2" in place
of "-z 2 -t 3". "-m 15" is experimental, but
seems to produce vastly better results than the
default, which is 10. "-M 3" is the default, but
since I'm experimenting, I like to have everything
specified explicitly so I'll remember to write it
down in my recording notes. (The right side of
every VCD and DVD I produce contain copious notes
on the recording and processing, so I can learn
from my previous work.)
The "yuvmedianfilter" command is there to lightly
blur the video, to remove spatial noise that
becomes really obvious once y4mdenoise has done
its job.
After all this, the video is sent to bzip2 and
mpeg2enc. It turns out that y4mdenoised video
compresses really well with bzip2. My experiments
suggest the compression ratio approaches DV
compression, i.e. 4-to-1 or better. Since I
don't know what -b/-q values mpeg2enc needs to do
my video properly, I like to have a copy of the
denoised video around. "yuv2lav" is NOT
sufficient for storing denoised video; the JPEG
compression introduces artifacts (especially on
slanted transitions between very light areas and
very dark areas), and seems to increase the
necessary MPEG bitrate by around 5%.
Finally, the video goes to mpeg2enc. I've
discovered that the -q parameter is very handy for
removing the last bits of noise that y4mdenoise
can't remove. As I lower the -q value, I
eventually find noise creeping back into the
video. I try to find a balance between detail and
noise, and for most of my y4mdenoised videotapes,
-q 3 is a good place to start. I'm hopeful that
the recent rash of mpeg2enc debugging will improve
this; as soon as I hear from Andrew Stevens that
he's looked at the 2 video clips I sent him & has
resolved those issues, I'll be back to testing the
CVS version of mpeg2enc. (Bugs are keeping me from
using "-q 2" or "-q 1" reliably.)
In the commented-out portions of my script, you
can see how I generate VideoCDs. After using
yuvmedianfilter to blur the video slightly (but in
a non-interlaced way, so as to lightly deinterlace
the video), I downscale it to VCD size, and before
sending it to mpeg2enc, I pipe it through
y4mdenoise with the error limit turned all the way
down. Here, y4mdenoise is not being used to
remove noise, it's being used to condition the
video stream numerically so that motion-detection
has a better chance of succeeding. It produces a
noticeable increase in detail on the final
VideoCD.
Hope this helps. Feel free to ask questions. I think
this is everything I know about producing
artifact-free
digital video, but who knows.
Steven Boswell
ulatekh at yahoo dot com
__________________________________
Do you Yahoo!?
Meet the all-new My Yahoo! - Try it today!
http://my.yahoo.com
makedvd.sh
Description: makedvd.sh
