Full DVD vs. D1 DVD for 8mm conversion

[tphaggerty]

I am converting a bunch of 8mm tapes using Canopus ADVC1394 for analog capture to DV (720x480DV) via Vegas 3, AVISynth->MainConcept for encoding, MF2 for authoring.

I am trying to retain a 1:1 ration for tape to DVD, which means I need to get 2 hours per DVD. That is not a real problem using VBR with a 4650 avg and 224 Layer 2 sound. So far so good (after a ton of experimenting and lurking on the boards here).

I have followed most of the "quality" type threads here, but none of them specifically answers this question. I have encoded at full DVD (720x480 at VBR4650) and D1 DVD (352 x 480 at the same VBR4650). Both provide the 2 hours and look very good, the same to my eye. When I use BitRate to evaluate the encoded streams, the D1 DVD has a significantly lower Q rating (~8.5 vs. ~10.8, which I understand means that it is a BETTER encode).

I want these DVDs for long term archival of the 8mm tapes, which are starting to deterioate (some are 8mm, some Hi8). Since the D1 DVD seems the same visually and seems to actually be a better encode according to BitRate (plus it encodes faster, not that big a deal), am I giving anything up (or gaining anything) by doing the D1 encode instead of the full DVD (given the source)?

[jfebus]

Hi,

In theory a clip using 352x480@4650 will have the same quality if a clip using 720x480@9300. This is due to the fact that you are using half of the resolution, giving you twice the bits for the encoding.

I have tried and I haven't see a diference between 352x480@2000 (4 hours) and 720x480@4000 (2 hours).

Regards,

[SaSi]

I would agree (in general) with jfebus comment. If you need to drop bitrate too low, it's better to decrease resolution, as it will help avoid pixelation in very active scenes.

My experience with half resolution encoding is that the image becomes too soft and slightly blurry, which I don't like. Of course, this is with DVD source streams, where both the resolution and the quality are high.

I would personaly opt to go for a lower average bitrate, such as 4800kbps, select a very low min bitrate, such as 200kbps, and a rather high max, such as 8000kbps. The min bitrate of 200 will allow the encoder to create reserves of bitrate for high motion scenes. If you then encode with 2-3 pass VBR with CCE, it will scan the movie, identify where it can encode with a very low bitrate and encode high motion scenes at a much higher bitrate than the average.

[tphaggerty]

The real question is whether the quality is BETTER if I encode D1 at the SAME bitrate as full DVD. In that case, the D1 encode will have the higher bitrate available and hopefully the encoder will use them (that is probably the real question).

Both of my test streams were at 4650 AVG, but 9250 max (using MPEG audio) and about 1800 min. The Max never went above about 5400 (not sure about the Min since BitRate doesn't show that). Like I said, BitRate said the Q factor was lower on the D1 encode, which in theory is better??

[ronnylov]

D1 and full resolution is the same thing, 720x480 in NTSC.
Half-D1 is 352x480 in NTSC. Just to avoid some confusion.

When the source is 8 mm or VHS then the "resolution" of the source is close to half-D1 and you won't see a difference (other than noise) at higher resolution. But Hi8 and SVHS resolution is somewhere between half-D1 and full D1 in detail level so in this case you'll loose some detail when using 352x480. But on the other hand it is harder to compress full resolution compared to half resolution and you need more bitrate to get the same visual quality att full resolution. That is why you get a lower Q-value in bitrate viewer (less compressed) at the same bitrate but a lower resolution. If you can't see any difference then choose the half D1 resolution if you prefer a clean video with less encoding artifacts but don't mind the risk of loosing some resolution on the Hi8 source. If you prefer details but can live with some artifacts (that may not be visible at playback) in heavy motion scenes then go for full resolution. Or use a higer bitrate and full D1 when encoding from Hi8 and use more discs.

By the way you can also choose the DVD resolution 704x480 instead of 720x480. 704x480 should be fine for Hi8 and it is a tiny bit easier to compress than 720x480. If you want a tiny bit more quality you can replace the area that are hidden in the TV-overscan by a black border. Resize to 672*448 and use a 16 pixel black border to get 704x480. This is easy if you use avisynth and generate a script with FitCD. If your source is home video then it's probably interlaced so you should use interlaced resizing and interlaced encoding. But this rezising and overscan optimizing is more for the advanced users.

[tphaggerty]

OK on the D1 vs. half-D1. The potential loss of resolution was what I was concerned about, seems like half-D1 might be the way to go for 8mm, D1 for Hi8. Lots of my video is outdoor stuff like water skiing and that seems to enduce some artifacts during encoding of fast scenes.

I am using AVISynth to cut out the overscan and adding back a border to center and keep the 720x480 size. I am using the MainConcept encoder to resize to half-D1, any thoughts on whether that is OK or better to do that in AVISynth?

If I use 704 (instead of 720), will that cause the video to be resized, or is it basically a crop? TMPGenc seems to default to 704x480, but I've always switched it back to 720x480 to avoid any resizing. You say that 704 is a bit easier to encode, is that just due to the smaller overall size of the stream, or ?? thanks, tphaggerty

[ronnylov]

Well, actually when resizing from 720 to 704 you should just crop. I don't think it's a big difference but less video pixels makes it a tiny bit easier to compress. I always do the rezising in avisynth with help from the program FitCD. It can align the picture to the macroblocks (16x16 pixel video blocks) that makes it easier for the encoder to encode the video. With avisynth I prefer the lanczosresize function because it creates a sharp and good looking resized picture. I also use denoising filters in avisynth. This helps the compression somewhat because if you can remove noise without loosing to much of the details then it's easier for the encoder to create a good picture. Convolution3D, deen, dust and Peachsmoother are some good denoising plugins worth testing in avisynth. To avoid loosing too much of the details I use the deoising filter before the resizing filter in avisynth. There are a lot of info on the doom9 avisynth forum.

I have never tested the mainconcept encoder so I can't tell if it's better in rezising compared to avisynth. I have tested Canopus Procoder and when using the mastering quality setting and 2-pass VBR this encoder can create good results also from difficult sources like home video. But it's really slow. I have heard that mainconcept is really fast but still good quality for DVD.

[DJRumpy]

CCIR-601 1/1 D1 (or cropped D1) is 704x480 (ntsc). CCIR-601 is 720x480. Half-D1 is 352x480 (I think I have the first two right anyway..I always get them wrong). Some refer to full D1 as 720x480. I think we just get lazy. Who wants to spit out CCIR-601 every time you refer to it? I have no idea where the 704 came from, as it wasn't part of the original CCIR-601 spec, which specified 720 pixel samlples. Maybe someone else knows?

Any reduction in resolution will improve quality due to bitrate, but at the cost of some clarity/sharpness. You should consider Half-D1 only if the bitrate you are encoding at is insufficient for the resolution.

704x480 is a good choice if your losing some quality in high motion scenes due to macroblocking. Resizing to 704 horizontal won't gain you much though.

You can always try a good temporal smoother, or 2-d filter. They can remove noise from your source. 2-D filters can smooth out images that don't have a lot of definition (like distant images). This will tend to make distant images a bit soft (blurry), while your main content is sharpened to a degree. That will reduce the bitrate requirement as well. I usually don't use a 2-d filter, as I find the blurring can be a bit strong at times. Again, it all depends on your source. It's it's not 'dirty', then don't bother.