Help w/ DV to DVD workflow

[ehickert]

I am in the process of converting a series of DV tapes (made with NTSC Sony TRV900 camcorder), and some analog Hi8 tapes to DVD-Video. The objective is to achieve the highest quality possible.

I plan to use the Sony camcorder's internal HW codec to digitize the Hi8 tapes to DV. (This has been suggested elsewhere in this forum, if you know of a better way, please let me know).

Once all the media is in the DV format, it will be edited in Adobe Premier on a Matrox RT2000 system and exported in the same native DV format.

I read, in this forum, that it will improve the compression process of DV to MPEG2 if you de-interlace and also do some noise filtering (or smoothing) of DV sources.

Avisynth seems to offer all these options. I have considered using plugins like: TomsMoComp(0,15,1) or Decomb FieldDeinterlace(blend=false). I am not sure which is best or which other noise filters should be used.

After coding the video to MPEG using a Cinema Craft Encoder SP 2.50 (depending on the Avisynth process, I suppose that the recommended settings for the CCE Codec might change, please let me know what you suggest here if you have any experience with this codec), the audio will also be processed in Sound Forge (to clean it up) and converted to AC3 (stereo) with BeSweet. The final stage would be to author the DVD with Maestro.

I have used Avisynth in a limited way (mostly following Doom9's how to guides for DVD backup. Any suggestions or comments on this whole process would be greatly appreciated.

[bb]

Your workflow sounds reasonable, but there are a few issues I would solve in a different way:

1. For DVD do not deinterlace. Better leave it interlaced, and use a higher bitrate instead. This will save you some time and artefacts.

2. You may use a little smoothing to get better results. There are a bunch of filters available for AviSynth; you may want to try Convolution3D with low settings.

3. For a better viewing experience for your originally analogue video you might want to add a little sharpening as the last filter in your chain. But that depends on your personal taste.

4. Although I would recommend to go the DV way as well you might want to try an analogue capture (e.g. through your graphics card's video in), too. Using the Huffyuv codec (or MJPEG) you might get better results, but probably not. The process will take longer this way. I think it's worth a try.

bb

[ehickert]

Thank you for your recommendations.

Do you think that the quality of the final DVD will be higher with the interlaced DV video versus de-interlacing?

[bb]

Yes, it will play smoother, and you won't get the typical deinterlacing artefacts and slight degradation in quality.

You can get better quality using progressive, i.e. deinterlaced video, if you are limited in bandwidth/bitrate. I recommend the progressive approach for SVCD having that 2.600 kbps limit (ca.), but for DVD, where you can raise the bitrate up to almost 10.000 kbps, you have more than enough to keep it interlaced.

bb

[ehickert]

Is there any other suggestions that you could recommend? other noise filters, etc?

Thank you for all your help.

[Xesdeeni]

If you can frameserve or export directly to your MPEG enocoder of choice without saving back to DV, you will also get better results (not to mention use less HD space and take less time). For DV footage that you don't touch, there won't be a difference. But for any that you edit, by adding a title, fades, transitions, etc., you will be re-compressing the frames, which introduces more artifacting. Feeding directly to your encoder will keep the video uncompressed. AVISynth has a plugin for Premiere if you want to use TMPGEnc. Cinema Craft has a plugin for Premiere, and in Premiere 6.5, the MainConcept MPEG encoder is built-in as an export.

Xesdeeni

[ehickert]

Thanks for your ideas. I think that would be the best way too.

I have been doing my tests with Convolution3D and really like it. Still playing around with the parameters to find a good balance.

I am not a programmer, but I wonder if it would be possible to use the Convolution3d avisynth filter in Premiere 6.5? How difficult would it be to convert it into a plugin premiere filter? This would make the workflow much simpler.

I am using the following script in avisynth with good results:

LoadPlugin("F:\SW\C3D\Convolution3D.dll")
LoadPlugin("F:\SW\monitor\avsmon.dll")
AviSource("F:\2000\dv001.avi")
ConvertToYUY2()
Convolution3D (0,4,4,4,4,2.8,0)
monitorfilter
ResampleAudio(44100)

I was thinking of also using the Msharpen filter at default settings, but I am not sure if this just works against everything that C3D has just done above.

Look for your comments

[ehickert]

Has anyone compared the Cinema Craft codecs to the MainConcept MPEG encoder? I have heard that they are comparable.

[bira]

ehickert,

Is there any problem applying convolution3d or any other filter on an interlaced source as DV? Wouldn't the filters mess around with the frames?

[ehickert]

I am not sure at this point. I am still doing tests with different material. I think the only real way to know is to burn a DVD and view it on a standard TV.

Maybe BB can comment.

[bb]

You should use Convolution3D with progressive frames. For interlaces sources, there's a little more work to do: First separate the fields, then apply Convolution3D, then weave the fields together again. This can be done with AviSynth's SeparateFields and Weave commands.

bb

[ehickert]

BB,

Could I ask you to clarify something?

Previously you suggested that I leave interlaced DV footage as interlaced if the final target would be DVD. Now you are suggesting to separate the fields, process them (Convolution3D) and finally weave them back together. Does this apply to DV footage that will be viewed on DVD video?

Just want to make sure I understand.

Thank you

[bb]

Clearly yes, for DVD stay interlaced. The SeparateFields / Weave workaround lets the progressive filter work on the (non-interlaced) fields. Then they are weaved back together, resulting in interlaced, but filtered, frames again.

bb

[ehickert]

bb,

I tried what you said and I now understand about separatefields and weave. Thank you.

Which sharpening filter would you use and with which settings?

[sh0dan]

Originally posted by bb
You should use Convolution3D with progressive frames. For interlaces sources, there's a little more work to do: First separate the fields, then apply Convolution3D, then weave the fields together again. This can be done with AviSynth's SeparateFields and Weave commands.

bb
Actually, since C3D is a temporal filter, the best way to filter would be:avisource("vid.avi")
separatefields()
vid_e=selecteven()
vid_o=selectodd()
vid_e=convolution3d(vid_e,....)
vid_o=convolution3d(vid_o,....)
interleave(vid_e,vid_o)
#insert non-temporal filters here.
weave()I personally wouldn't use sharpening filters, but you could look at Unfilter and Blur2. Look at Avisynth.org
Edit: Also try msharp from Donald - it also produces good results.
edit: script spelling

[bb]

Hmm, that's a matter of taste. You should try some filters on your own and decide what you like best. I usually don't sharpen at all, but I don't produce DVDs (yet), only SVCD or MPEG-4.

There are some AviSynth filters floating around. I recommend to use one of them instead of VirtualDub for performance reasons.

bb

[ehickert]

Thanks for the additional suggestions. I will try them and share the results with you.

[bb]

You're welcome.

bb

[bira]

I'm back! Thanks for the answers!

Do I have to use "separatefields" in order to use "FixBrokenChromaUpsampling" ?

FixBrokenChromaUpsampling is not temporal, right?

Thanks

[bira]

This is what I am doing:


AviSource("file.avi")
separatefields()
vid_e=selecteven()
vid_o=selectodd()
vid_e=convolution3d(vid_e,0,4,4,4,4,2.8,0)
vid_o=convolution3d(vid_o,0,4,4,4,4,2.8,0)
interleave(vid_e,vid_o)

FixBrokenChromaUpsampling

weave()

ResampleAudio(44100)

Is it ok?

[sh0dan]

No, FixBrokenChromaUpsampling isn't temporal, so you can put it anywhere. Your script seems fine! How's the results?

(It also isn't spatial, so in fact you're completely free to place it). Perhaps I should have written:

#insert spatial filters here

and the

#insert color correcting filters here

at the bottom

[bira]

Convolution3d works great! Video is less noisy and detail is not lost.

But, I am getting different results when using FixBrokenChromaUpsampling. Is there a sample video so I can test it?