TV Caps - What Do You Do?

[Piper]

For tv captures, I most often capture at 352x480 (NTSC) YUY2 using Huffyuv. I've found very little or no quality gains capturing at higher resolutiosn.

For IVTC sources, I use GreedyHMA which I've had good success with, esp with Star Trek Original Series. The delemna comes when I capture a pure interlaced or hybrid source and my destination is a 352x480 DVD which I want to be able to watch on both my TV and on my PC.

I've used GreedyHMA, DeComb, FieldDeinterlace with every setting I can think of/get my hands on. The playback is always either jerky, or
the picture loses too much detail. I realise there's no magic setting, but there must be a balance somewhere.

I've scoured the forums and found scattering examples here and there, and based my experiments on them so I have an idea what the right balance is, but I'd be interested to see what you're currently using to deinterlace tv captures. Please post here!

Thanks in advance

[vidiot]

...just a "quick ´n dirty" reply:

If 352 x 480 works for you qualitywise and the main problem is interlacing (and IVTC) -> try going to capture in vcd resolution
(352 x 240).
O.k. you capture just one field then, but maybe you´ll be surprised
what quality you can achieve (I feel the whole "capture crew" looking angry at me now...:D ).

It all depends on your source and of course on your hardware.
The jerkyness is maybe introduced through a field order problem!?
Did you already check this?
Or maybe just shifted fields...
Or both...

I guess many people would like you to make a short clip available for dl...
Then one can help.

On the other hand -> This might not be the best forum to ask for help
with tv caps...

regards
Harald

[ronnylov]

I don't deinterlace when converting to DVD, I encode it as interlaced. But I don't watch them on the computer other than of testing purposes, I mainly watch it on TV.

But you can set up your DVD playing software to deinterlace during playback or perhaps using some directshow filters that can deinterlace at playback. Perhaps its possible to use graphedit and make a filtergraph for playback. I never tried this myself.

If you still want to deinterlace a pure interlaced source before encoding you may also want to try the avisynth filter TomsMoComp.

[Piper]

Thanks for the replys.

I have tried reducing the resolution to VCD using ReduceVerticalBy2 and the quality isn't bad, but it IS less than the source and that can't be helped. I've generally found there are no quality gains when capturing higher than 352x480 either. A side effect is, 352x480 is a valid DVD resolution (as is 352x24) - but I prefer to keep it at 352x480. Mind you, one could fit many more minutes on a DVD at 352x240. :-)

In regards to jerkyness - how do I check if there is a field order problem, or shifted fields, or both?

The best solution I have come up with so far is while experimenting with GreedyHMA. I've put together the following settings based on suggestions and examples I've seen here.

GreedyHMA(1,0,1,0,1,0,60,90)

This works quite well, but there's an unwanted side effect where it appears the color is being downsampled (ex 16-bit vs 24-bit). I'm not sure what part of GreedyHMA is doing this. I've tried numerous variations on the above line, but I honestly don't know enough about what each value does to tweak it much further. Anyone?

Perhaps when I'm back at my home PC I'll post a couple of screenshots to illustrate this effect.

Thanks

[vidiot]

Originally posted by Piper

In regards to jerkyness - how do I check if there is a field order problem, or shifted fields, or both?
Simple:
Open your video with Virtualdub and also get smartdeinterlacer (2.7 beta 2).
(You´ll get it from Donald Grafts site)

There you´ve every option you´ll need to check everything mentioned.
The final check is always your tv!

[Guest]

SeparateFields() is an indispensable Avisynth internal filter for diagnosing field problems. Simon's ViewFields() is also very useful.

[McQuaid]

Like what other people have said, your capturing resolution is fine.
Don't start capturing in 352x240 even if your encoding to vcd. You need the full frame to properly ivtc/deinterlace the source.

If you do capture 352x240 you'll notice on telecined source material that it'll seem like a frame repeats every 5th frame. This is because of the telecine pattern where a field would be repeating, but you have just captured it as a frame. So in this case it's still beneficial doing a simple ivtc for the dup frame detection.

If the source is true interlaced material and you ultimately want to watch it in dvd format, do nothing to it. It won't look good on a pc, but you'll be keeping all the motion in the 60 fields. Otherwise deinterlace.

The problems crop up with hybrid material i.e. film and true interlaced sources. Some shows that do this are star trek tng and Oz. If someone could confirm this I think star trek tng is a special case, because it's not just film and interlaced sources, but sometimes interlaced special effects are laid on top of telecined material. This makes it difficult if not impossible to ivtc correctly. If this is true, probably the best method again is do nothing to it, if your burning to dvd. Otherwise decomb has some methods of dealing with hybrid material.

Btw, I have never encountered progressive material in my captures, but just the other day I encountered 2. Curb your enthusiasm and a nirvana video I captured. I was surprised looking through the source there was not one interlaced frame.

I've read some DV cameras are capable of 30 fps progressive, but was kinda surprised of a 10 year old video having this as well.

[scmccarthy]

So in this case it's still beneficial doing a simple ivtc for the dup frame detection. In that case, it would not be a full IVTC, but only decimate(cycle=5)

Stephen

[^^-+I4004+-^^]

>I've read some DV cameras are capable of 30 fps progressive, but was kinda surprised of a 10 year old video having this as well.

why?
there are loads of music videos shot on film....
in the past even more than today.......
you can distinguish video from film just by
looking at the color tones of imag.....
film looks more natural,video looks washed-out...

[McQuaid]

The reason I was surprised was because it was 30 progessive fps.

I admit, I don't have much knowledge of traditional film cameras, so maybe it's trivial to just shoot at 30 vs 24 fps.


Btw, which DV cameras do this? I believe that I read this is only in the higher end models. Something that backs this, I was reading an article recently where the director Robert Rodriguez is shooting a new film and wants to keep the budget really low. He is using miniDV cameras for the shoot and decided on PAL cameras as the 25 fps is closest to film.

It would be nice if digital cameras could shoot in ntsc,pal,30/24 progessive. I would think this wouldn't be that difficult being digital cameras, but maybe this is more difficult to achieve with digital cameras.

[Piper]

A friend of mine has a 3 year old Sony "progressive" digital camcorder. I have my doubts whether it's truly progressive given the price he paid, but I could be wrong. I do know it tends to produce very dark images. My Panasonic DV Camcorder produces pure interlaced footage. Encoding it straight to MPEG-2 at full 720x480 resolution with little or no filtering produces great results. Again, for DVD's which are destined for set top boxes rather than a PC, it's perfect.

fyi: I was never questioning my capture resolution - I never capture at 352x240. I only meant to say that I find no quality gains in capturing (tv) at resolutions above 352x480.

I had another look at my GreedyHMA settings. What I mistook in my earlier post for colour downsampling is actually blending during motion when it attempts to combine fields. When looking at individual screen shots, motion areas of the image look as if the colour has been downsampled to 256 colours.

I've played with SeparateFields and Weave a bit and while I understand what they do - I don't quite follow how that helps with deinterlacing. Yes, I know....www.100fps.com :-)

[vidiot]

I don´t want to hitchhike(?) this thread,
but one thing you should think about on 25 or 30 fps progressive cameras (whether they´re DV or other type):
The only advantage is that you have Frames which doens´t need deinterlacing (ok I admit: BIG advantage sometimes)

But here´s the bad point:
Think of sport games - especially ball games (soccer, tennis ect.)
- Got the point?
How far does a tennis ball with ~140 km/h go during one frame?

So interlacing just gives you the half field, but better control...

... and yes the next update on DV camcorders will be 50 (60) fps progressive, don´t you think?

Harald

[McQuaid]

Yes, interlacing has it's advantages. Your capturing 60 moments in time (although fields vs frames), so your going to have more motion in the recording.

But interlacing is something that needs to be dealt with when encoding to alot of video formats. Although obviously there are expections like svcd and dvd natively supporting interlacing.

BTW, I read that one of the HDTV formats is 60 frames progressive. I would think it's going to take us roughly double the amount to then store the same amount of video. I doubt this will apply to film (although it should be applied there as well) but capturing television shows will be handled differently.

Regarding you not seeing a benefit from 352x240 captures. Even if you find that to be perceptibly true, the benefit of capturing full frame (480 lines) is undeniable beacause of deinterlacing or doing a proper ivtc.

With deinterlacing as was mentioned about being smoother because it was recorded with 60 moments in time, when you deinterlace your taking into account that motion instead of discarding it at capture (by in effect only capturing 30 moments in time).

And even more so with doing ivtc on film source material (movies, lots of cartoons). The benefit here is two-fold. Your returning the capture to its original frame rate. So you'll be encoding 24 fps instead of 30 (yes 23.976 instead of 29.97, but I hate having to point that out). The first benefit is, because there are actually less frames per second, the encoder can either dedicate roughly 20% more bits per frame, or achieve the same quality in roughly 20% smaller file size.

Many formats can gain from this. Obviously mpeg4 has no problem with this frame rate, vcd has native support for 24 fps, and dvd and svcd has the ability to do whats called 3:2 pulldown on playback. This brings the frame rate back up to 30 fps on the fly, instead of having the frames encoded this way.

Which brings me to the second benefit, from what I understand, it's more difficult for any encoder to encode interlaced/telecined frames as this appears more random, instead of pure progressive frames.

Now there is a drawback, these telecined frames, the frames created from doing the telecine procedure, give an illusion of in between motion and can make a video appear smoother. But again, these frames are not really true unique frames. From my personal experience the loss of the this 'faked' motion is worth the tradeoff of having pure progressive frames.

Capture pretty much any cartoon full frame, and do one by either leaving it alone or deinterlacing, and the other by doing ivtc on the source. With cartoons especially, the almost ghosting effect of these interlacing is so glaring to me that it reuins the capture.

I only capture full frame (480 lines) if I want to keep it, I occasionly capture half frame just to watch something and delete it later.

Bottom line. If your not capturing full frame, your not capturing at all. :)

[Piper]

Bottom line. If your not capturing full frame, your not capturing at all.

I agree 100%. That's what I was trying to communicate. Always always always capture at full frame (480 lines). What I don't see any point in is capturing ABOVE 352x480.

I'd like to retain my 352x480 capture with as much detail without resorting to reducing it to 352x240 - though that does have its advantages - being automatically deinterlaced. You are throwing away detail at that resolution however... I run this debate with myself daily...

More experimenting with GreedyHMA tonight and I now believe I've found the right combination for my video (not film) captures is GreedyHMA(0,0,0,0,1,0,0,0) while retaining full 352x480 resolution.


My script I'm using now for video tv captures is:
GreedyHMA(0,0,0,0,1,0,0,0)
Convolution3D(0,3,5,2,4,2.8,0)
MSharpen(threshold=20,strength=40)
ConvertToRGB() <-- TMPGEnc

For film sources like Star Trek original series I use:
GreedyHMA(1,0,4,0,1,0,0,0) <-- Nice & fluid motion
Convolution3D(0,3,5,2,4,2.8,0)
MSharpen(threshold=20,strength=40)
ConvertToRGB() <-- TMPGEnc

I've been struggling with this for ages - sometimes it just helps to talk it out. (watch me change my mind tommorow) :-)

Thanks for listening

[McQuaid]

Ok, I see what your saying about capturing above 352x480 and not being able to see any visual benefit.

Actually, I do all my captures at 352x480. I guess the main reason for this is capturing at full frame is much more significant than capturing above 352. I think a lot of people would be hard pressed in picking out a capture done at 352x480 and resizing to 640x480 vs a capture done at actually 640x480. Obviously, there is more visual information at 640 but it's visually negligble. What you have to take into consideration is your final encoded resolution. If your going to encode at 640x480, then hard drive space permitting, cap at that resolution. But if your final resolution is 352x480, cap at that resolution. Since dvd supports 352x480, I personally would capture at that, as I agree that the difference between 352 vs 640 is not as much visually vs how much more you can store on disk.

But what you said about reducing to 352x240 and the advantage being that it's automatically deinterlaced. This not correct. It will appear to be deinterlaced, but all you've done is discard half the visual information. True deinterlacing, will take into account the full interlaced frame and deal with the combing effect of interlacing while trying to retain the motion.

[scmccarthy]

Does full frame really mean 352x480? I would have assumed full frame means 704x480 or 720x480. Can you capture at 704x240? I bet that would look sharp on a computer at 30fps progressive. Although I don't think TVs have that much resolution per line.

Stephen

[McQuaid]

Full frame is defined as both fields making up the frame.
For ntsc 480 lines is full frame, I always forget pal, 525 I think. But the thing here is we know the exact amount of lines that makes up a television picture, but you can't say exactly how many vertical lines there are. When people use 640 vs 352, it's because this is generally believed to be good value to approximate how much visual information is in one horizontal line. But again it's an approximation. It's kinda ling film. Noone can really say film needs x by y resolution to retain the exact amount of information in the original photo. Well, same thing here but in the case we at least know the exact value of y.

You would be disgarding much more significant data when capping at 704x240 vs 352x480 even though storage wise they would be roughly the same. The reason because with the former you would be definately discarding half the information (and half the motion with interlaced source), but with the latter, you'd just be using a lower approximation of the information in the horizontal range.

[Ookami]

http://forum.doom9.org/showthread.php?s=&threadid=32575

http://forum.doom9.org/showthread.php?s=&threadid=34122

[aviaggio]

By no means am I an expert. I'm learning as I go along. But here's what I've learned so far:

In general you don't need to do ANYTHING if you've capped from TV at 29.97 and you're encoding for DVD. No IVTC, no deinterlacing, NADA. The footage is already in the right format for TV -- 29.97 fps, 3:2 pulldowned if necessary, and interlaced. Why mess with it?

Now if you were encoding to MPEG-1 or DivX, then it's another story. 24 fps film material that you capped (you can tell by seeing sequences of 3 progressive frames and 2 interlaced frames in your capture during hi-motion sequences)needs to be IVTCd and can be encoded at either 24 fps or 29.97 fps (see below). Native 29.97 material should just be deinterlaced.

Mixed material is a bitch. What you do really depends upon the content. If one portion is film and the other video and it doesn't alternate, I'd encode them as separate ranges and then join the files later. If it's all back and forth, but primarily film, then IVTC, knowing the video material might be a bit jumpy. If it's mostly video, just deinterlace, knowing you might have some smearing or bits of artifacting in the film material. You need to do some test clips and see what works best.

You can, if you choose, IVTC film caps even if your intent is DVD. The advantage to this (at the expense of added encoding time) is that you may get a higher-quality image in CBR or smaller file sizes in VBR because you'll be encoding only 24 fps instead of 29.97. The trick to this is correctly setting up TMPEGEnc set to encode the video.

Have TMPEGEnc IVTC your cap. Then set your source to progressive and use "3:2 pulldown when playback" as the Encode method, and "23.976 fps/29.97 internally" as the frame rate. IVTCing reconstructs a 24 fps progressive source which is then encoded with instructions for the player to do 3:2 pulldown on the fly. Commercial DVDs made from 24 fps material are done this same way.

Again, this is what has given me good results.

Tony

[joshyg2]

I always capture at the following parameters:

Compressor: Huffyuv
Video Resolution: 768x576
Frame-Rate: 25
Audio Frequency: 44100hz
Channels: Stereo

After capturing I closely examine the frames of the captured material, if one of them is progressive then usually it is Telecided, so I try using a number of techniques of getting around that, by using decomb and GreedyHMA.

If the source is 100% Interlaced, then I encode to Interlaced is either DVD or SVCD. If I am doing a series I encode to DivX and then use FieldDeInterlace or Telecine().

If the source is Progressive (usualy episodes), (with NO combing) I just encode as Progressive without doing any other processing (except for Noise Red, and Resizing).

I captuere off Analog, so to get rid of noise(evil) I use PixieDust.

When captureing of VHS, I use my JVC S-VHS VCR, with TBC (Time Based Corrector) turned on, and then i use flaxen's VHS filter to align the colors again.

Josh

[Si]

Instead of Flaxen's VirtualDub filter you might want to try my ChromaShift filter because it works in YUY2 colourspace and could save you a colourspace conversion.

regards
Simon

[Piper]

ChromaShift has done wonders with some of my old VHS sources. I highly recommend it.

[Chainmax]

Is ChromaShift really better than flaXen's stabilizer? Where can I find a good guide on it?

[Si]

Is ChromaShift really better than flaXen's stabilizer?

Well Flaxens's operates in RGB colourspace and he shifts I and Q relative to Y (search Google for YIQ stuff)

Mines works in YUV and shifts U and V relative to Y.

So sometimes mine could be better as it gives a slightly different effect and no colourspace conversion.

Where can I find a good guide on it?

The best guide is available via my site :D

(OK OK - its the only one ;) )

regards
Simon