DV vs lossless capture of VHS

[themaster1]

if you want to examine-compare noise use this at the end of your script:
Histogram(mode="luma")

[Brad]

Thanks to lossless & DV samples posted by FLP347 in a thread about capture cards, I now have a real-world home movie clip that meets most of the requirements of the OP. It's interlaced PAL, so it's more relevant to him than an NTSC sample and the associated endless debates about whether 4:1:1 is an issue for consumer analog tape conversions.

Only thing is, it's Video8, not VHS. The maximum horizontal resolution is 272 TVL according to the source I'm reading (Sencore Tech Tips), compared to 250 TVL for VHS HQ & Beta I (Wikipedia). 272 seems bizarrely specific given the margin of error in this subjective measurement! I haven't been able to find any other sources who give that number; others vary from 240-270. I think part of the confusion is due to the existence of the little-used Video8 XR.

In any event, I think the differences found in recording/playback equipment within each format are probably greater than the theoretical differences between the formats themselves. So I figure it's more than close enough. But I did want to note it for fellow pedants.

With FLP347's original uploads and my encodes, I have the lossless file and three different DV versions to share. All of the DVs show blocking not present in the lossless source, regardless of whether encoded by the hardware codec in the Sony Digital8 camcorder, Canopus' DV software codec, or Cedocida's DV software codec.

Now my question is, what should I be targeting to show whether there's any visible effect in the final delivery version? The OP suggested he would encode each clip for DVD to look for differences. I'm personally more interested in showing how they compare when encoded for viewing with x264.

Does anyone have any suggestions as to CRF/bitrate/etc.? I'm not sure what people would consider a generally "realistic" scenario. The degradation of the DV-intermediate version is obvious in stills if CRF 10 is used, but I suspect most people aren't really using a CRF that low for their viewing copies.

[poisondeathray]

Thanks vaporeon800,

If you're bored, why not post several versions ?

I think still, realistically, DVD-Video would (still) be the most common scenario for VHS or Video8 for a final format. Typically bitrates might be 5-8 Mbps

For x264, the current "typical" CRF range would be 16-20, so why not 18 ?

I guess the question is "how bad" is the additional blocking / degredation? I'm going to guess only very perceptive people will notice the difference under normal playback conditions

You mentioned 2 software + 1 hardware, but I'm mildly curious to see if there are significant advantages to other DV encoders ? e.g. Maybe Mainconcept or Avid etc....

[jagabo]

Keep in mind that PAL DV has 4:2:0 chroma, not 4:1:1 like NTSC DV.

[lordsmurf]

Nobody ever has issues with PAL DV. It's the 4:1:1 NTSC DV that has issues. It cooks the colors quality. As others have noticed, there are oddities. You need to remember that DV was never intended for conversion -- only shooting. The only company to use it for conversion was Canopus (and the clone products like DataVideo).

Lossless is easily better.
MPEG is better as well -- especially high bitrate non-DVD specs.

When you have a choice, don't use DV. When you're screwed (example: using a Mac for capturing), you'll just have to live with the quality loss.

I've never understood why some people try to argue all this. Accept that it is what it is, and plan your workflow accordingly.

As far as noise goes, that's the device. And Canopus boxes tend to introduce mosquito noise (digital grain).
In terms of block noise, that's a DV codec issue. (All the DV codecs have it to some degree.)

I find Avisynth scripts to "fix" DV to be silly. Just avoid it entirely. (And I would argue that you can "fix" it anyway -- information is tossed out on conversion.)

4:2:2 > 4:1:1 > 4:2:0 is never pretty.

I was having these conversations 15 years ago.

[Brad]

It's the 4:1:1 NTSC DV that has issues. It cooks the colors quality.

4:1:1 is more than sufficient for the chroma resolution of all consumer analog tape; jagabo demonstrated this for VHS in another thread in which you participated. If NTSC DV "cooks" the colors, it could be due to heavier compression on the chroma channels or the specific algorithms used.

I would like to see an example of what you mean by this terminology. A few others use it as well.

(In a thread on DigitalFAQ you seemed to agree with me that 4:1:1 itself is not the problem, and suggested that the 4:1:1 -> 4:2:0 conversion is the culprit.)

The only company to use it for conversion was Canopus (and the clone products like DataVideo).

Sony used DV for playback conversion of Video8/Hi8 in at least one model of each generation of Digital8 camcorders, starting with the DCR-TRV103 in 1999 and ending with the DCR-TRV480 in 2005. (Models from 2000 added the ability for live passthrough from the analog inputs to digital.) I think their usage predates the ADVC boxes, doesn't it?

[lordsmurf]

The ADVC boxes have been out forever. They were made for Pentium III computers.

Given that shooting 4:1:1 is fine, I still agree that other factors are at play. But the downconvert of the colors during a conversion must play some hand in the problems that are seen. And the 2>0>1 conversion makes it worse, yes.

Explaining why it's bad is not as important as just understanding that it is. I care more for the results.

[jagabo]

Changing chroma subsampling is essentially resizing the chroma. We all know that you should avoid resizing when not necessary. And DCT compression always results in ringing artifacts. So it should also be avoided when possible.

But sometimes you have to work with what you have.

[pippas]

But sometimes you have to work with what you have.

I use a combination of Panasonic NV HD90 VHS recorder as the tape player, via SCART to a Panasonic DMR E55 DVD recorder, and then via S-video to my Canopus ADVC50. I'm very pleased with the results, working with 'what I have'...

Mind you, I am in PAL land, so that may help a bit?.....

[lordsmurf]

Mind you, I am in PAL land, so that may help a bit?.....

Yes. PAL is 4:2:0, and not really an issue.
NTSC 4:1:1 is the problem.

[jagabo]

4:1:1 chroma subsampling isn't much of a problem with NTSC VHS caps if handled correctly. You need to use a DV decoder (or colorspace converter) that interpolates chroma rather than duplicating chroma when resampling.

[ron spencer]

-30-

-31-

[lordsmurf]

-31-

-32-

[Clockwork]

-31-

-32-

-33-

[pippas]

-31-

-32-

-33-

-34-

I suspect this could soon get very silly.......

[Brad]

You mentioned 2 software + 1 hardware, but I'm mildly curious to see if there are significant advantages to other DV encoders ? e.g. Maybe Mainconcept or Avid etc....

I only have one of those, and not on this machine. Perhaps someone else can volunteer to grab the clip and encode it.

Further inspection reveals that this clip doesn't scream "blocking" in the way I thought. I had been looking at it using point-resizing, which of course makes the hard edges of the blocks plain. With a resizer someone might actually use, the artifacts are a lot more subtle.

Given that, should I still post each video encode? Here are screenshots with LanczosResize(width*2,height*2). I used SeparateFields rather than deinterlacing, to avoid any interplay there.

Lossless


Sony DCR-TRV238E DV (hardware)
Lossless -> Canopus DV (software)
Lossless -> Cedocida DV (software)


Lossless -> Cedocida DV -> interlaced x264 (CRF 18)
Lossless -> interlaced x264 (CRF 18)


Encoding this to DVD specs seems to be a lost cause, unless my HCEnc settings are horrible. Or maybe it's as simple as denoising first, I dunno.

[johnmeyer]

The Canopus DV looks pretty close to the original lossless. For video of this quality, I would find it hard to justify the extra time and disk space to encode to lossless.

I have both the Sony DV encoder and Mainconcept DV encoder. I guess I also have the Canopus encoder that someone gave me in payment for a job ten years ago. From discussions back then, these three were considered way beyond the Windows DV encoder, which was absolutely awful. They were also judged to be much better than almost any other DV encoder.

Do you want me to encode from the original lossless still image, or is there a link to a video clip that I am supposed to use. I'd be happy to post my results.

[Skiller]

I'm inclined to play a little with that clip as well.
Would love a lossless sample if possible.

[Brad]

Here it is in Ut Video form.

EDIT: And the Cedocida DV encode from lossless.

[ron spencer]

-31-

-32-

-33-

-34-

I suspect this could soon get very silly.......

-10,000,000,000,000,000,000,000,000,000,000,000,
000,000,000,000,000,000,000,000,000,000,000,
000,000,000,000,000,000,000,000,000,000,000-

[johnmeyer]

I guess I'll have to back out on doing test encodes. Nothing I have will read the UT Video codec. Normally I can convert using ffmpeg, but the resulting uncompressed file wouldn't play either. I can certainly handle Lagarith, HuffYUV, and I wouldn't mind installing MagicYUV. I would think that any of those would work just as well for this test, although perhaps they don't handle the color depth you want.

[Brad]

I had simply used the one that's smallest and most convenient for me, since no preference was specified. Lagarith and Huffyuv are both larger and slower.

Here's the Lagarith version.

[newpball]

I've never understood why some people try to argue all this.

I suspect it is bit phobia.

They see very large files and get uncomfortable, and their first thought is - compress!

[johnmeyer]

I had simply used the one that's smallest and most convenient for me, since no preference was specified ... Here's the Lagarith version.

OK, that works great. Here are two links to the resulting encodes. For the first file, I used the excellent Sony DV codec that is built-into Vegas. For the second, I used the MainConcept DV codec. I kept the field order the same (upper field first). There are definitely a few minor artifacts, but if you can download the still images and zoom way in, I think you'll find that the Sony DV codec is really, really good. This is what I've been saying here for some time. I think that several people who have been trashing the "DV codec" here in this forum only know the really crummy versions of that codec. Not all DV codecs are created equal.

Here are links to the two versions:

Sony_Vegas_DV.avi

MainConcept_DV.avi

Here are snapshots of the same field (which I bobbed) from the original and then the two encodes:

Lossless


Sony DV


MainConcept DV

[pippas]

. I kept the field order the same (upper field first).

I thought DV was always lower field first?......

[poisondeathray]

Thanks for the tests guys

Just some observations on the screenshots only, havent looked at the actual videos yet - but it looks like Mainconcept's DV is prone to macroblocking e.g around the foreground soldier's arm, the pillar . Sony's DV looks like it drops details almost like a NR was applied, blurring shadows like the lady with the white sweater's pants have the details obliterated

Normally lagarith would offer better compression than ut video if they used the same color space. I'm wondering if something went wrong because the lagarith file looks to be larger

[johnmeyer]

I thought DV was always lower field first?......

Nope. It is happy to do it either way. I'm not sure if it would play correctly if you put it back on a DV tape and played it from your camera, but the codec itself can handle it either way.

In general I always recommend not changing field order when encoding. In theory the encoder should be able to handle that without problems, but I proved, many years ago, that Sony Vegas Pro really screws up when you simultaneously re-size and also change field order. The problem is not field reversal, where you get that funky "vibrating" look to motion, but simply a reduction in spatial details. It is probably a problem deep inside of Vegas, and not something specific to a codec, but I try not to tempt fate.

And, just to complete the thought, you can encode DVDs with either upper or lower field first, and the player and TV will play it just fine.

[poisondeathray]

The ut video sample was 4:2:2 . I didn't download the lagarith sample, but I assume it was 4:2:0 ? That would explain the size relationship. What was the original, original capture ?

Also, the ut sample was BFF

[2Bdecided]

Encoding this to DVD specs seems to be a lost cause, unless my HCEnc settings are horrible. Or maybe it's as simple as denoising first, I dunno.

Anyone who can see any difference between DV and lossless is going to be utterly utterly horrified at DVD-compliant encoding of challenging interlaced video content, viewed with the same level of criticism.

Cheers,
David.

[johnmeyer]

... but it looks like Mainconcept's DV is prone to macroblocking e.g around the foreground soldier's arm, the pillar . Sony's DV looks like it drops details almost like a NR was applied, blurring shadows like the lady with the white sweater's pants have the details obliterated ...

I agree about the macroblocking, although it is only evident in some sections of the still shot, mostly around that woman's white shirt.

I don't see any detail reduction in the Sony DV, other than what is caused by the macroblocking. Here is a link directly to the images, so you can download them directly:

Lossless.png

Sony DV.png

MainConcept_DV.png