restorative usage for deinterlacing VHS

[lordsmurf]

I would actually suggest this be done at the encode phase (like TMPGENC), not the capture phase, but it applies to restoring video captures...

REMEMBER! In general, DO NOT DEINTERLACE... this is ONE EXCEPTION to the rule

I had a fairly bad clip to work with last night... it had lots of jitter that showed up horrible on screen... made the interlace lines bounce.... so I decided to drop half the data... in other words, a de-interlace ... I ran the EVEN ADAPTIVE in TMPGENC PLUS from an AVI capture...

The results were decent... though VHS deinterlacing normally leaves undesirable artifacts by dropping half the visual data.. this time it helped the video by dropping the jitter to half...

It was the lesser of two evil... jitter or deinterlace stair-step artifacts? ... I chose the latter... the jitter gives me a headache... the deinterlace I'll just suffer through, but at least it won't leave me noxious...

Just wanted to share...

[trevlac]

REMEMBER! In general, DO NOT DEINTERLACE... this is ONE EXCEPTION to the rule

... though VHS deinterlacing normally leaves undesirable artifacts by dropping half the visual data..

I don't really want to start a vicious debate, although comments are always welcome.

For the record: I never deinterlace captures. All of my souce is interlaced. I mainly save cartoons (not much else worth saving that's not on DVD anyway), and my limited experience has show deinterlacing to be a bad thing.

However
Some very smart people have put a huge amount of effort into figuring out how to deinterlace well. Check this out : http://neuron2.net/journal/journal.html

And a few things I think I know (may be wrong always still learning).

1) MPEG does a much better job at compressing progressive material. The offset in the fields causes problems in the motion estimates. The guys who wrote procoder may have figured this out. I havn't bought a copy yet. (I'd like to try it for my miniDV stuff).

2) 4:2:0 Color should be an even bigger problem because (from my limited understanding), the color from one of the fields is interpolated from the other field. Again the time offset is not a good thing here.

3) deInterlacing does not throw away 1/2 of the data. The challange is in the motion areas where the field offset is appearent. AKA, a proper deinterlace of a static picture is the same picture. And please don't think video is not static pictures. Mpeg compression is based on the idea of only saving the changes. So to get a 20:1 compression means there is a lot of 'non movement'.

Conclusion

Understanding deinterlacing is not for newbies. I do not deinterlace my stuff (i am still learning much). I appreciate LS pointing out there are cases to deinterlace. There seem to be more cases, but I don't know.

Cheers,

Trev

[lordsmurf]

Interlacing is a one way street. Once you go there, you cannot reverse the process. It's like shredding a document. Once you've done that, nothing you can do will restore it to original condition. Sure, you can tape it together again, but it's not the same. It is visibly scarred.

Deinterlacing of interlaced source normally causes jerkiness and jaggy edges on diagonal shapes and curves. The only real exceptions are IVTC, but that's not going to always work.

Yes, it is a collection of images, but when interlaced, they share a symbiotic relationship, and separating them will either kill them or at least make them exist in a less than perfect environment (forced progression).

If I had it my way, everything would be progressive. Thanks to lazy tv engineers from back in the day, that just never happened. I wandered around for months when I first got my ATI AIW card, wondering why my captures looked like garbage while real DVDs did not. The reason was because nobody ever explains interlace and it's finite universe.

Actually MPEG-2 was optimized over MPEG-1 for usage with interlace specifically for broadcast and storage (DVD). So 4:2:0 and interlace should not affect the quality. At most, it'll change the filesizes, bitrate allocation and encode time. And I believe the motion estimation is done from field-based prospective, which is one reason to properly let the encoder know the source format.

General rules are keep interlace from interlaced source, especially if destined viewer is interlaced (tv set). Progressive if source was progressive or deinterlace only if final format is computer only (normally streaming web video) and on-the-fly deinterlace programs (WinDVD,PowerDVD) are not an option.

[trevlac]

@LS

Thanks for your comments. I made a major typo in my prior post. I do not deinterlace my stuff.

However, I am "still wandering around" trying to learn more.

The real mystery to me is why Donald Graft (and others) put so much time into deinterlacing. Maybe they just watch it all on a PC?

Anyway, thanks again. I respect your firm conclusions. As a newbie, I'm just not there yet.

Trev

[trevlac]

Here is some of why I have doubts ....

Is digital TV always deinterlaced at some point before you get it? Maybe it is re-interlaced too.... Clearly if it comes out of an analog wire from your receiver it is interlaced because that is the standard.

To quote a technical source:

"De-interlacing is required any time an interlaced source is format converted, regardless of whether the destination is progressive or interlaced. De-interlacing is required in interlace to interlace conversions in order to provide more temporally aligned samples for the resampling process."

Is a quote from http://www.broadcastpapers.com/sigdis/motion01.htm

Another doc I've been trying to fit the context of is : http://www.teranex.com/support/docs/Issues%20DTV%20Upconversion%201.2.pdf

That is a discussion of HDTV, but it appears they take the view that they can deinterlace without loosing resolution. Maybe only for HD, but the 1st link implys that even SD digital is deinterlaced some time in the process. I can't say that is they way I get it from Dish.....

But, as you see, I'm not sure on this topic.

Clearly deinterlacing allows for source to be viewed on digital TVs. Is it possible to do it and still have it look good on a SD TV?

One conclusion I would cautiously make is that high end studios face this issue with HD and SD digital, and that the standard "hobby" info on the net is just one side of the coin.

[lordsmurf]

You can speed up the ~60 images per second (NTSC) to restore the progression, then resize on the fly. This is how an ATI card preview works. Separates the fields but shows the images twice as fast.