PAL/NTSC Conversion Guide updates + de-blur restoration (add Avisynth use)
 

Since we have a few new* members here that are skilled at Avisynth, I thought I'd acquire a bit of community input before the guide gets an update (and/or a companion) in January. (*"New" compared to the guide's writing.)

I refer to the guide for Encoding/Converting > PAL/NTSC Standards from http://www.digitalFAQ.com/guides/vid...t-pal-ntsc.htm

Specifically, there two areas to focus on, where updates are desired:
- deinterlacing options
- de-blurring poorly converted videos (that are your source before you re-convert it back)

At the bottom of the guide, you'll find two sample files, clips from the Australian DVD release of Inspector Gadget (which has been fully released down in Oz, but has not been released in North America). There is a good bit of frame blending, which creates the ghosting/blurring effect.

Deinterlace improvements:

When the guide was written, VirtualDub was not yet to version 1.9.x, and did not yet have the Yadif deinterlacer. That is generally preferable to available Deinterlace Area-Based filter for VirtualDub, although not always.

So that's a GUI option, for those who want to avoid the "complex" scripting and plugins for Avisynth.

A better approach would undoubtedly be to use a combination filter set in Avisynth, using this script:
(Note: YES, I prefer ffvideosource, as mpeg2dec's d2v files are an extra step, and are not necessarily any more accurate than ffmpeg, based on my past experience and observations. This is more of a preference issue, and I opt for the one that is easier and still looks the same. I would, however, include alternatives, as a footnote link.)

This is a fairly easy script.

De-blurring / De-ghosting additions:

What I think could be improved on further is to attempt to remove the ghosts caused by the original poor conversion, by use of Avisynth's srestore() filter. And this is where I'd most like some input on quality script examples for the guide. It is one of the better documented scripts in the Avisynth wiki, but it's still lacking in examples for the visual learners of the world (which includes me).

Audio conversions:

Since the writing of the guide, Sound Forge 9 also came out, which allows for more direct input (AC3), and from what I was reading a few days ago, now includes frame-accurate measurements for time stretch, as opposed to that useless percentage method.

and then...

Finally, it may be split across a few pages, in steps, instead of one long page.

Thoughts? :)

And if this works well, we may collaborate on several guides.

My use of Avisynth is messy. I gather bits and pieces from here and there and it changes from clip to clip. Primary source of information and examples has been doom9 forum. What comes to mind..

For PAL/NTSC/PAL I think I've used capable the mvtools2 package http://www.avisynth.org.ru/mvtools/mvtools2.html - examples in the bottom.

About the use of yadifmod and NNEDI2 - NNEDI3 is out and can drop-in replace NNEDI2 - http://forum.doom9.org/showthread.php?t=147695 Quick test I ran on i7 920: Yadifmod+NNEDI2 runs at 25~fps and Yadifmod+NNEDI3 at 35~fps (PAL VHS 576i50 -> 576p25).

"order" and "field" parameters could be set to -1 (uses avisynth's internal parity value) and then declare parity beforehand using AssumeTFF() or AssumeBFF(). Forcing it to TFF will cause problems to people handling BFF material.

srestore is easy. Let's say your from US and have a PAL sourced video. The conversion is usually done with blended frames, which you call blurry. This can be perfectly undone.
Will return the original, 25fps PAL video, progressive.
You can even reconvert it back to ntsc in a different way by slowing it to 23.976fps and using pullup. This won't show the blurriness anymore but will play slower.

You can tell these kinds of conversions by stepping through the video field by field, and note that there are 'ghost' frames where say two copies of an arm in motion appear on the same field, and one is dimmer than the other. This is called a blend.

In the example you see a ghost of the arm above. When unblended by using surrounding frames, you get the 2nd picture. This shows the unblended arm in the upper position. There is of course also the unblended frame with the lower position.