Log in

View Full Version : PAL Deinterlacing or not & how + Avisynth


S”nTė£
13th September 2005, 15:50
Situation: Working on backups of my DVD's, PAL interlaced (according to DGindex (http://users.skynet.be/TailwinD/DGindex.png)). Going for HQ anamorphic ~1.1gig x264+AAC MKV encodes for personal use only (archiving 4 movies per DVD, 4x1.1GB fits nicely). The thing is, it's only going to be displayed on computer CRT's and LCD's -progressive displays- don't have any real TV's in the house anymore, main movie watching is on a HTPC+LCD.

Currently using the very very default AVS:
MPEG2Source("e:\dvd.d2v")
Trim(7024,0)
Crop(0,80,0,-80)
UnFilter(-5,-5)

I've spent the entire day doing nothing but reading posts and guides so yes, I've searched for the obvious before posting this. What I would still like to add to the script is some basic, not too extensive denoising. I was thinking about using both unfilter and undot as the spatial denoisers (temporal denoisers I'll look at later). Undot requires the source to be progressive. Now, I already found the best way to go about this from here (http://forum.doom9.org/showthread.php?t=97628):

LeakKernelBob(order=0, threshold=7)
Undot()
assumebff().SeparateFields().selectevery(4,0,3).Weave() #reinterlace

Info() & DGindex tell me the frame order of the source is top field first, so I reckon I have to change order to 1 and replace assumebff() to assumetff()? Now what I also could not make out:

Most posts I read suggest to keep the source interlaced hence the "reinterlacing" step, but most of these posts come from topics where the ultimate goal is conversion to an interlaced medium (DV->DVD for example). Keeping in mind the target MKV medium & computer display output, is it still better to keep interlacing? Also, does x264 even do interlaced encoding, I've read in some thread that the developers are not big fans of this. But doing some test sample encodes with the codec without doing some obvious deinterlacing (just the above avs script in virtualdub on fast recompress + vfw codec) goes well. Yes, I know that for this last bit this is not the right forum...

--> If going for interlaced, are the parameters of selectevery still 4,0,3 considering that I'm in PAL-land? If going for progressive, I assume I can safely drop the last line and replace LeakKernelBob with LeakKernelDeint (since LeakKernelBob doubles the framerate/count to 50fps)? Since I'm going for quality here, is TomsMoComp better in the deinterlacing department without dropping the speed too much?

off-topic: highlighted the questions to filter them from the side-info. Sorry about the nick btw, I made the account over 1.5 years ago but never posted. Now I wouldn't dream of using such a "1337" characters :/. Finally, if this is the wrong forum, just tell me and i'll try getting it moved.

scharfis_brain
13th September 2005, 16:14
combs visible during movements?

YES: interlaced treatment needed.
NO: no interlaced treatment needed. so, forget about ALL of this stuff.

S”nTė£
13th September 2005, 16:20
No combs visible whatsoever, so I'll guess I'll forget about making it progressive; thanks for answering that part
BUT to use undot I think I still need an answer on:
*so I reckon I have to change order to 1 and replace assumebff() to assumetff()?
*are the parameters of selectevery still 4,0,3 considering that I'm in PAL-land?
since the source need to be 'bobbed' for that one.

Many thanks.

stephanV
13th September 2005, 16:44
if your source is not interlaced it effectively has no fields so you can remove

assumebff().SeparateFields().selectevery(4,0,3).Weave() #reinterlace
as well

Forgive my rudeness, but: do you really need all of this filtering? Or are you just doing it because other people are doing it too?

S”nTė£
13th September 2005, 17:22
Well DGindex tells me it's interlaced so I assume my questions still stand.

As you were guessing, the source is pretty clean, but the reason I'm looking into filtering is to increase the compressibility, not to increase visual splendor. That's why I'm using filters which to the eye make almost no difference, but should matter to the codec so that it can use the saved bits for scenes that really need it. In the end I'll only have unfilter and undot, the fastest denoisers according to the tests, so "all this filtering" is not really that much filtering. I just want to do a one-time filterchain setup that I can use on most DVD's.

There's some truth in your statement, I'm just pretty new to this and want to learn as much as I can.

There's one issue that I like to resolve tho, it's a shimmering pixel effect on static backgrounds. I'm guessing I'll have to use temporal denoisers for them but would like answers to my open questions first before I begin doing my 'research' on them.

jggimi
13th September 2005, 17:50
Well DGindex tells me it's interlaced so I assume my questions still stand...See the PAL section in www.doom9.org/ivtc-tut.htm for more information. The tutorial is a little out-of-date, but the discussion on DVD2AVI (now known as DGIndex) and its reading of MPEG-2 flags should be helpful.

Boulder
13th September 2005, 17:51
Well DGindex tells me it's interlaced so I assume my questions still stand.

As scharfi told you, you don't need to do any interlaced treatment. You can't make the source progressive because it already is ;) DGIndex can only tell how the stream is flagged, you must trust your own eyes to make the decision. It's pretty much normal to see progressive content encoded as interlaced, the other way around it will ruin the video.

S”nTė£
13th September 2005, 18:06
Ahaaaa, now that clears up a bunch and puts scharfi's post in perspective, thx a lot you 2! Now I can move on ^^