MPEG-2 vs. DV

[moontrip]

Hi, everyone. I've been reading a lot of the posts on this website, and doing some additional research, but I haven't been able to find a clear-cut answer to my question, so here goes...

I'm converting a lot of old analog video (mostly VHS) to digital for archival purposes. In order to capture the highest quality video possible, should I encode it as MPEG-2 or as DV? (I don't have the storage to capture uncompressed video, so I'm trying to decide between these two.)

I will probably burn DVDs from the video after the conversion, but that is not the primary purpose of capturing the video. I want the video captured at the highest quality possible for archiving purposes. I know the technical differences between MPEG-2 and DV compression, but I can't find a lot of opinions as to which is better quality.

Thanks in advance for your help!

Bill

[SatStorm]

For PAL users, it is DV hands down

For NTSC users, is questionable, if your source is ultra high and you use a hi-end capture card. If not, it is again DV

Most users I know, are extremely happy with DV capture (combined with a good VCR and a TBC)

[laspis59]

For the best quality you can possible achieve,capture with a lossless codec (No quality loss whatsoever)

Huffyuv is one of the fastest and most popular lossless codecs.

[moontrip]

Thanks for the replies. So you say DV is better quality than MPEG-2. So from an overall quality standpoint, does the larger data stream make up for the reduced color space? Is their a recognizable difference in color quality between the two? Thanks again.

Bill

[thecoalman]

https://www.videohelp.com/forum/viewtopic.php?t=257651

As stated above it depends on your source. The example in that link are from a very high quality source, VHS capture won't even come close.

Persoanlly it's my thought that discs are cheap, I'd go with the DV or as suggested above the Huffy codec.

[edDV]

First you need to realize how low quality VHS is even in SP mode. Broadcast NTSC is ~ 4.2 MHz (340 horizontal lines of resolution and 486 vertical lines) luminance and ~ 1.0 MHz for chroma components.

For VHS, luminance is filtered well below 3 MHz (~240 horizontal lines of resolution max). The amount lower depends on the recording VCR quality. Typical cheap home recorders use very cheap filters that probablly roll off around 2-2.5 MHz (~200 horizontal lines of resolution). Chroma is modulated in the color under process at only 500 KHz per component.

S-VHS allows luminance to be recorded out to ~4MHz (320 lines of horizontal resolution) in SP mode but chroma remains at only ~500-700 KHz per component.

Quality issues for VHS capture divide into two broad issues: capture method and capture format.

Capture Method:

Capture method is mostly a hardware issue. VHS is recorded to tape with separate Y (luminance) and C (chominance) components. A Y/C (S-Video) connection will allow the separate components to be sent to the NTSC/PAL decoder and then to the A/D. A composite connection will mix the components forcing a filtering process in the capture card to reseparate them. That may reduce quality and introduces artifacts. It is more an issue for S-VHS which has high detail luminance and chrominance in the same frequency space forcing the need for a comb filter.

The typical A/D converter is only 8 bits. Serious archival captures would be done at 9, 10, 12 or even 16 bits to give digital filters more quantization levels to work with. Modern capture cards sample at CCIR-601 (aka D1) reference or 13.5 MHz or "half D1" 6.75 MHz. or somewhere inbetween into YUV color space (either 4:4:4, 4:2:2 or 4:1:1 ratios).

Without getting into details, any of these sample rates easily handle VHS bandwidth, but increased quantization bits (over 8bit, 256 levels) are desirable for post digital processing. Unfortunately consumer capture devices easily oversample VHS but none offer > 8 bit quantization. 10bit cards are still very expensive. 10 bit+ quantization would allow better levels processing in the digital domain before encoding to 8bit DVD MPeg2.

8 bit quantization encourages analog levels processing (proc amp) before A/D. VHS horizontal jitter can also be compensated with a timebase corrector (TBC) before A/D for best results.

Capture format:

Again all formats are captured to CCIR-601 referenced sample rates (13.5 MHz luminance, 6.75 MHz Cb and Cr) typically at 8bit. This describes an uncompressed capture at 8bit D1 4:2:2 and requires about 75GB for one hour of capture. As said above, this oversamples the VHS but restricts quantization vs 10 or more bits. It can be argued that a true "half D1" (i.e. 352x480/576, 6.75MHz Y and 3.37MHz Cb/Cr aka 2:1:1) would be more than enough sample rate for VHS and would consume 38GB/hr. uncompressed.

75GB/hr is a bit much so compression becomes a priority. Huffyuv gets "lossless" rates down to ~28GB/hr, DV (4:1:1) compresses 5:1 for 13.5GB/hr. and DVD MPeg2 (4:2:0) compresses ~15-25x for 2-8GB/hr.

So, serious "archiving" should be analog pre-processed or >8bit. Archive formats could be "uncompressed", Huffyuv, D1, DV, MPeg2 (various compressions) or to DVD-5 at 4:2:0 at various compression rates. Quality roughly follows that order for those seeking minimal loss. Remember VHS is very low quality to begin, so all this effort needs to be justified. I would only worry about home VHS and 8mm camcorder "family" tape archiving at this level of quality.

I would argue a 10bit 2:1:1 (~45GB/hr) capture card to 10bit 2:1:1 MPeg2 (2-10x compression) would be the best archive for VHS but hardware does not exist at consumer prices.

[vhelp]

Capturing lossless YUY2/UYVY etc. is obviously out of the question
for your process, per your first post. It would cost your approx
1.2gig / min of HDD space if you were to use the format above.

Though Huffy will compress the above YUY2 (via its codec implementation)
further. I haven't tried ALL codecs out there, but I have had my fill.

DV is even smaller than the HUFFY codec compression, but reduces
some chroma information during the sampling stage, (in addition to
its similarity with MPEG's DCT method of compression) though it is is
not noticable to the eye.

Regarding your MPEG-2 vs. DV ...

Based on your final choice of codec for your capturing, and then
finally, archiving to either MPEG-2 or DV, I would say that you
would be better off with choosing DV as your final archive. With
MPEG-2, you are subject to its own set of limitation, NOT TO MENTION
your own limitation in your skills/knowledge/techniques, etc etc.

So, DV would be the better choice for archival purpposes, but then
you have another problem. Size per media to archive video to.

For one hours worth of video would require:
** 4.3gig --- MPEG-2
** 13.5gig -- DV (would need 3 dvd-r disks per hour)

If you are trying to retain as much source detail as possible, and
given your limitation requirements, I would suggest to first capture
as lossless as possible. That means, something like the codecs that
use YUY2422 or UYVY422 for the final video.

This will give you the most detail to retain in your next phase.

Then, depending on your final choice, MPEG-2 or DV.

Perhaps if you run some tests, you could size up the decision.
I don't know.. I'm just suggesting further. Anyways.

You have to come to a comprimise with the two.

-vhelp 3625

[edDV]

I would agree with vhelp except I see no advantage to 4:2:2 over 4:1:1 sampling. Most basic capture cards are unable to capture 4:1:1 so you may have to use the extra HDD space for 4:2:2 only for that reason.

DV capture devices (4:1:1 NTSC or 4:2:0 PAL) or DV camcorder Pass-Through (4:1:1 NTSC or 4:2:0 PAL) should give equal chroma quality*.

Capture to 4:2:0 has other issues that I will skip since it won't make any difference for VHS sources.

* 4:1:1 samples analog U and V at 3.375 MHz. Since VHS (and 8mm) only contain 500-700KHz actual chroma bandwith, the Nyquist Law requirement for sampling >2x is easily achieved.

[moontrip]

edDV, the coalman, and vhelp, thank you, thank you, THANK YOU for your replies! I can't tell you how much I appreciate the depth of your detailed responses. It only confirms my feelings that this forum is the best out there on the subject of digital video, because of people like you (and I've read other postings of yours, edDV, that are equally helpful and insightful).

Anyway, as far as the information you have provided, I think the consensus is that DV is indeed better than MPEG-2, as far as my situation is concerned. edDV, to your point about the video justifying the lengths you'll go to archive it, I should add that these aren't home videos. This is a professional setting, and I do have access to professional equipment. However, we've got hundreds of hours of analog video to digitize, so space (and time) is an issue.

One of my concerns about DV had been based on things I had read that said transcoding DV to MPEG-2 (to make a DVD) can result in lots of artifacts. But coalman, you seem to have anticipated my question with the link you've provided. And if I'm reading the results of your tests correctly, it sounds like, contrary to what I've read, DV to MPEG is actually preferable to MPEG to MPEG. Right?

One issue I have is that I'm working in the Apple universe, and surprisingly I have now learned that there are virtually no MPEG encoding cards available for the Mac. Really, my only choices are DV or uncompressed. So now I'm thinking I might go with a pro DVD recorder for the bulk of the video, and capture the important stuff either uncompressed or in DV. If I go uncompressed, I'm assuming (hoping) that the Quicktime codec is as good as the Huffyuv codec that everyone keeps talking about here.

Anyway, thanks everyone, I really appreciate it. This site rocks!

Bill

[davideck]

The typical A/D converter is only 8 bits. Serious archival captures would be done at 9, 10, 12 or even 16 bits to give digital filters more quantization levels to work with.

Capturing at 8-bit resolution does not limit the associated digital filtering to 8 bits. 12 to 16 bit digital processing after 8 bit A/D conversion is typical. The additional bits represent fractional levels (1/2, 1/4, 1/8, etc.) to maintain the accuracy of intermediate calculations before the final result is rounded back to 8 bits.

An 8-bit A/D conversion of a VHS source surely results in at least one LSB of noise. Therefore, I don't see how VHS captures would benefit from additional quantization resolution; the additional bits would just be quantizing the source noise...

[thecoalman]

And if I'm reading the results of your tests correctly, it sounds like, contrary to what I've read, DV to MPEG is actually preferable to MPEG to MPEG. Right?

Absolutely, using a higher quality source is always preferable which my link above clearly shows. Keep in mind that clip was hand picked for that test and the bitrates were well below what you would use for mpeg at that resolution.

I had originally posted that on another forum, a poster there was insisting that mpeg capture was just as good as the original DV. That may be true for lower quality source but once you get into the realm of high quality source the difference becomes more evident.

[edDV]

The typical A/D converter is only 8 bits. Serious archival captures would be done at 9, 10, 12 or even 16 bits to give digital filters more quantization levels to work with.

Capturing at 8-bit resolution does not limit the associated digital filtering to 8 bits. 12 to 16 bit digital processing after 8 bit A/D conversion is typical. The additional bits represent fractional levels (1/2, 1/4, 1/8, etc.) to maintain the accuracy of intermediate calculations before the final result is rounded back to 8 bits.

An 8-bit A/D conversion of a VHS source surely results in at least one LSB of noise. Therefore, I don't see how VHS captures would benefit from additional quantization resolution; the additional bits would just be quantizing the source noise...

Post processing does generate multiplication products out to 10-12 bits agreed but input quantization sets overall pecision.

the additional bits would just be quantizing the source noise...

This may be true for limiting application to VHS. Noise is anywhere from -20dB to mid 40's best case (with internal VCR DNR). The opportunity comes from attacking noise that repeats field to field.

Here is my thinking and this has been confirmed by broadcaster experience. Original CCIR-601 was 8bits. This was fine for digitally generated video but placing a CCIR-601 device in a A/D, D/A chain revealed servere problems with the need to proc-amp analog levels prior to every A/D. 16-235 gives about 9% overhead for levels but real world levels are skewed to white (hot and clipped) or low. Low levels suffer from insufficient quantization. 8 bit video only has 219 levels (with 9% room for superwhite spikes) for a perfect signal. A signal that is 10% low for gain has only 175 levels of quantization and so on.

Current broadcaster practice is to use 10 bits with level 64 for black and 940 for white. White headroom remains 9% but "soft clamp" techniques are used during A/D to manage overshoots, but low levels can be digital processed (proc-amp) with much better precision with 876 quantization levels.

VHS has greater "levels problems" than just setting white and black. Digital noise reduction, gamma correction, RF byproduct removal and other sophisticated digital filters could work better with increased amplitude quantization. It may even be possible to do some post A/D timebase correction if 10-12 bit quantization is combined with oversampling.

[laspis59]

Sorry to go off topic but edDV,christ you have totally confused me.
I can see that you know a LOT about anything to do with video.

[edDV]

That was just a side issue with davideck

[edDV]

And if I'm reading the results of your tests correctly, it sounds like, contrary to what I've read, DV to MPEG is actually preferable to MPEG to MPEG. Right?

Absolutely, using a higher quality source is always preferable which my link above clearly shows. Keep in mind that clip was hand picked for that test and the bitrates were well below what you would use for mpeg at that resolution.

DV has much higher luminance quality* than DVD MPeg2 luminance but whether this makes much difference for realtime VHS capture needs testing.

DV has 5:1 intraframe compression (similar to JPeg) and no interframe compression.

[davideck]

the additional bits would just be quantizing the source noise...

This may be true for limiting application to VHS.

The context of my statement was VHS captures, as was the premise of your Capture Method discussion.
We seem to agree; 8 bit A/D conversion is fine for VHS.

So I was basically challenging this statement...

I would argue a 10bit 2:1:1 (~45GB/hr) capture card to 10bit 2:1:1 MPeg2 (2-10x compression) would be the best archive for VHS but hardware does not exist at consumer prices.

[edDV]

the additional bits would just be quantizing the source noise...

This may be true for limiting application to VHS.

The context of my statement was VHS captures, as was the premise of your Capture Method discussion.
We seem to agree; 8 bit A/D conversion is fine for VHS.

So I was basically challenging this statement...

I would argue a 10bit 2:1:1 (~45GB/hr) capture card to 10bit 2:1:1 MPeg2 (2-10x compression) would be the best archive for VHS but hardware does not exist at consumer prices.

I guess where we differ is I hold out for better future noise reduction and DSP technology and for this reason I would ideally archive at 10bit noise and all.

My vision is that at a minimum, "digital proc-amp" and maybe digital domain TBC will be possible when combiled with advanced noise reduction. That is without the quantization effects of 8 bit capture (219 active levels).

[davideck]

I would argue a 10bit 2:1:1 (~45GB/hr) capture card to 10bit 2:1:1 MPeg2 (2-10x compression) would be the best archive for VHS but hardware does not exist at consumer prices.

I guess where we differ is I hold out for better future noise reduction and DSP technology and for this reason I would ideally archive at 10bit noise and all.

If that is your hope, then why 2:1:1?

My vision is that at a minimum, "digital proc-amp" and maybe digital domain TBC will be possible when combiled with advanced noise reduction. That is without the quantization effects of 8 bit capture (219 active levels).

Then perhaps 4:1:1 or 4:2:2 would be more ideal.

[edDV]

I would argue a 10bit 2:1:1 (~45GB/hr) capture card to 10bit 2:1:1 MPeg2 (2-10x compression) would be the best archive for VHS but hardware does not exist at consumer prices.

I guess where we differ is I hold out for better future noise reduction and DSP technology and for this reason I would ideally archive at 10bit noise and all.

If that is your hope, then why 2:1:1?

My vision is that at a minimum, "digital proc-amp" and maybe digital domain TBC will be possible when combiled with advanced noise reduction. That is without the quantization effects of 8 bit capture (219 active levels).

Then perhaps 4:1:1 or 4:2:2 would be more ideal.

But only to support the "TBC" luminance temporal resolution issue. Chroma only has ~80 samples per line. 2:1:1 is more than adequate for sampling VHS bandwidth. It's generally accepted that 3:1:1 is enough for sampling Betacam SP.

[davideck]

2:1:1 is more than adequate for sampling VHS bandwidth.

And 8 bits is more than adequate for sampling VHS S/N ratio.

That's why capturing at 10 bit quantization and 2:1:1 bandwidth seems out of balance to me.
I would suggest the following formats;

8 bit 2:1:1
8 bit 4:1:1
8 bit 4:2:2
10 bit 4:2:2

If I was concerned about future technology enough to quantize VHS to 10 bits, I would also sample at 4:2:2 for all the reasons that you have mentioned.

[edDV]

If I was concerned about future technology enough to quantize VHS to 10 bits, I would also sample at 4:2:2 for all the reasons that you have mentioned.

Of course. I'd like to see capture cards go to 10bit A/D 4:2:2 at reasonable prices. I'd settle for 9bits A/D in with digital proc amp rounded down to 8bits 16-235.

[anitract]

Well I know how much everyone loves DV around here lately (and really I am just thinking of getting into it due to me wanting a new camcorder), but for me the file-size advantage of mpg2 is enough to make that my preferred method for **VHS** captures now. As was said before, VHS is pretty low-grade to begin with, so I do not feel that I am losing any detail what-so-ever with mpg2.

[davideck]

I'd settle for 9bits A/D in with digital proc amp rounded down to 8bits 16-235.

I would argue that substituting an 8 bit A/D into this configuration would yield the same results.