DV vs lossless capture of VHS

[2Bdecided]

Can anyone provide...

A) a short clip from a raw lossless capture of a VHS tape, and
B) the AVIsynth script they used to "restore" that lossless capture


...please?


I want to see, really, if capturing it using DV-AVI rather than lossless would have hurt the process at all.

I propose to run the restoration once on the lossless version, once on a DV-AVI copy of the lossless version, and post the results. I'll also encode the results for DVD.

The B) part is optional, if that helps. I can make something up or do nothing if need be.
The capture A) is preferably from a home movie, using an S-VHS VCR with a line-TBC - but anything will do if that's too difficult.

(I don't have any way of capturing lossless, and I've always been happy with DV-AVI. However, people keep telling me lossless is the way to go, so I want to see if I can see the difference!).

Many thanks,
David.

[jagabo]

I don't think you'll see much difference doing it the way you propose. The bigger issue is that many DV devices have no proc amp controls. So if you get crushed brights and darks there's nothing you can do.

[2Bdecided]

Hi jagabo,

I know that (from bitter experience once!) - but crushed blacks and whites are easily detected in a capture, so it's easy to diagnose that as a fault. Whereas people keep saying DV causes other damage too. I want to see, once and for all, if those of us capturing VHS to DV could be doing noticeably better. As I don't have a capture card that will do lossless, I'm asking if anyone can provide a sample.

It won't be a fair test, because I can't pass it back through my ADVC110 - but I can at least pass it through a DV codec, which isn't too unfair since the ADVC110 seems to contain a decent enough DV codec. The capture part of the ADVC110 appears to be accurate but unforgiving, so a decent lossless capture shouldn't be too unfair a comparison to that part.

Cheers,
David.

[jagabo]

Someone posted a Lagarith encoded file here this morning:

https://forum.videohelp.com/threads/360802-Clean-up-darker-horizontal-lines-from-VHS?p=...=1#post2286862

I don't know if it's suitable or not. And it's PAL so it will be YUV 4:2:0 in DV, not YUV 4:1:1. I suppose you could crop it down to 720x480 and AssumeFPS(30000,1001) then make new Lagarith and DV encoded videos. I think a professionally shot movie would be a better source. Something you could inverse telecine.

[sanlyn]

I don't understand the argument here. The posted sample was captured (apparently) to lossless YUY2 from PAL VHS. Get that same source and player and capture it through some device to DV, if that's what the debate is about. Unfortunately the original source, according to the owner, no longer exists.

[poisondeathray]

David is in PAL land, interested in PAL VHS capture - why would a PAL source not be appropriate ?

[jagabo]

I don't understand the argument here.

He want's a lossless YUY2 capture to convert to DV. That simulates a case where you capture with a DV device that is otherwise exactly the same as the YUY2 capture device. Ie, the only difference is the 4:2:2 vs 4:1:1 chroma subsampling, and the DCT compression.

[sanlyn]

No it doesn't. Get a VHS source and capture it to lossless AVI (and not with a $20 capture stick). Get the same VHS source and capture it to DV. How is a lossless capture converted to lossy DV supposed to be an improvement?

[poisondeathray]

I don't understand the argument here.

He want's a lossless YUY2 capture to convert to DV. That simulates a case where you capture with a DV device that is otherwise exactly the same as the YUY2 capture device. Ie, the only difference is the 4:2:2 vs 4:1:1 chroma subsampling, and the DCT compression.

Huh ? That sample is a PAL 4:2:2 lagarith capture , just encode that with PAL DV (4:2:0) and it should "simulate" what you could expect if he had captured it in the first place PAL DV . What's this NTSC stuff ?

No it doesn't. Get a VHS source and capture it to lossless AVI (and not with a $20 capture stick). Get the same VHS source and capture it to DV. How is a lossless capture converted to lossy DV supposed to be an improvement?

It's not supposed to improve it. It's a simulation test - supposed to show the difference of capturing in lossless vs. DV. ie. "How much do you leave on the table" by capturing in DV instead of lossless . Or worded differently: what is the benefit or gain of capturing in lossless 4:2:2 vs. DV in the PAL 4:2:0 case .

[jagabo]

No it doesn't. Get a VHS source and capture it to lossless AVI (and not with a $20 capture stick). Get the same VHS source and capture it to DV.

The two devices will be using different A/D chips, proc amps, AGC, sharpening filters, etc. That makes it difficult to determine what differences are caused by DV compression vs. the other uncontrolled parameters.

How is a lossless capture converted to lossy DV supposed to be an improvement?

It simulates a case where a DV device has the exact same capture hardware and processing as the YUY2 capture device -- the only difference being the DV compression. Otherwise you can only argue that a particular DV capture device is bad for VHS capture, not that all DV capture devices are inherently bad for VHS capture.

Of course, 2bdecided only has one (?) DV capture device so what he really needs to know is how his device performs relative to some good YUY2 capture device. He won't know until he does the definitive experiment of capturing the same source with both devices.

[sanlyn]

Capturing the same source with both devices is what I refer to. All my captures are with ATI AIW 7500 or 9600XT cards.

True, if you capture from VHS with a good card and player, then use that same player (or, say a camera as a player) and capture that to DV -- which is not the same thing as copying DV to DV -- you would indeed have different processors, etc. The DV processing might be different with with various DV devices, and some would use different codecs. You might then be able to say that so-and-so DV process was worse4 than or better than another one. But in any case, even with a very clean DV capture, you are left with the need to do any cleanup in lossless AVI and then to re-encode to something other than DV.

I'd have to say that if DV is to be your delivery format, it would make little sense to capture VHS to lossless (unless there's some heavy cleanup to do) and then encode it to DV.

[Brad]

I captured a cartoon VHS some months ago in an attempt to provide some data one way or the other regarding capturing to various codecs. I found that my sample was "too good" to show any significant differences until I got down to MPEG-2 or XviD at medium-low bitrates, so there wasn't any point in posting it and I put it on the back burner until I find something more appropriately degraded.

Is it correct to say that two standards-compliant DV codecs, presented with the same input, may produce different outputs (like MPEG)? If so, it's tough to extrapolate from a lossless capture using a certain ADC converted to DV using (say) Cedocida vs a hardware device using a different ADC and a DV hardware codec.

If the question is about the color subsampling rather than the compression, I don't see any way that 4:1:1 would look worse for VHS unless something went horribly wrong in the downsampling. I couldn't even make it look worse for LD, where 4:2:0 is insufficient to reproduce highly saturated color transitions. Lordsmurf claims the hardware has issues: http://www.digitalfaq.com/forum/video-capture/5156-advice-proposed-video-2.html#post26615

[2Bdecided]

Someone posted a Lagarith encoded file here this morning:

https://forum.videohelp.com/threads/360802-Clean-up-darker-horizontal-lines-from-VHS?p=...=1#post2286862

Thanks jagabo - I hadn't thought of trying that one because (as a home movie, rather than due to the technical faults) it's unwatchable. I guess it'll do for a technical test.

You've made me realise though that there must be old threads on here and doom9 which have the perfect test material for this. Since no one has posted a favourite clip to this thread, I'll go trawling.

Cheers,
David.

[2Bdecided]

This isn't what I was talking about, but may be of interest...

I pushed my ADVC110 to its limits by creating a test card with the full range of possible Y, U and V values. The TIFFs and AVIsynth script are attached, as is an excerpt of the MPEG-2 encoding of it that I burned to DVD.

Many of these YUV combinations create illegal RGB colours which PCs just can't handle. However, early camcorders recorded just such colours to tape, and CRTs display them with far more graceful overload than a PC. That's what prompted me to do this test, because I need to capture and manipulate video with these illegal colours.

At the time I did this test (five years ago), the DVD players I had around all output the full YUV range via composite and almost the full range via S-Video (some didn't do blacker than black, and one didn't do whiter than white). On RGB SCART output they clipped the RGB range far less (i.e. reproduced a wider colour range) than a PC will, but didn't preserve as many illegal colours as composite and S-video outputs. If you think about the DAC ranges on these analogue outputs, you'll see why: YUV 0-255 gives voltages that the Y/C DACs can handle when the results are illegal RGB colours, whereas of course when outputting the RGB values themselves, some of these are way outside the voltage range of the RGB DACs.


Viewed on a CRT, the very ends of some of these chroma ramps merged together on some TVs, while all discrete steps were visible on others. The colours were hopelessly oversaturated of course, but were always the right hue. In contrast, the RGB clipping of illegal colours which occurs in PCs gives you completely the wrong hue.

The ADVC110 happily grabbed the full range, and pretty accurately too it seems (though the DVD player I used is an unknown quantity here). Composite and S-video captures are attached. Be careful how you view them - these test patterns will make any colour upsampling errors rather obvious.


Anyone who thinks DV devices can't capture luma or chroma levels correctly should probably reconsider their opinion in the light of this evidence.


The chroma is blurred horizontally and vertically far more than the source 4:2:0 (YV12) digital video. The additional horizontal blur is entirely down to the limited chroma bandwidth of analogue video, and was present on the output of the DVD player. The additional vertical blur is partly due to the PAL filtering, and (I assume) partly due to the DV chroma sub-sampling. Remember that, while you're probably examining this on a progressive PC display, this was sent as interlaced video, so both the PAL filter and the DV subsampling stretch over 3 lines.


If you need good horizontal chroma sharpness (for NTSC), or good vertical chroma sharpness (for PAL), then DV is a bad choice. VHS exhibits neither of these (PAL VHS recordings have typically gone through two or more vertically blurring PAL filters), so my belief is that DV doesn't really hurt VHS.

But we shall see...

Cheers,
David.

P.S. I wouldn't use the ADVC110 for capturing composite. It doesn't seem to have a comb filter. My HV20 faired better when I tried it, but as I'm not trying to capturing composite that was as far as I went.

[2Bdecided]

I'd have to say that if DV is to be your delivery format, it would make little sense to capture VHS to lossless (unless there's some heavy cleanup to do) and then encode it to DV.

I assume the delivery format is always MPEG-something, always 4:2:0, and either interlaced SD, progressive SD, or upscaled to HD.

The only use I have beyond this is screen-capping frames to print to photos. If they're close-up shots of people's faces, the results aren't as bad as you might expect, but otherwise I really wouldn't try this with a VHS source.



Can anyone think of another use for a VHS capture that doesn't end up lossy encoded and delivered with 4:2:0 chroma?


Cheers,
David.

[2Bdecided]

No it doesn't. Get a VHS source and capture it to lossless AVI (and not with a $20 capture stick). Get the same VHS source and capture it to DV.

The two devices will be using different A/D chips, proc amps, AGC, sharpening filters, etc. That makes it difficult to determine what differences are caused by DV compression vs. the other uncontrolled parameters.

Exactly. That's why I'm trying this test, which leaves DV codec vs lossless as the only variable.

Obviously if all available analogue>DV capture devices have bad ACGs, bad sharpening filters, etc, then you can't use them for that reason. However, my belief is that the ADVC110 captures S-video sources pretty faithfully.


Cheers,
David.
P.S. I'm not trying to sell anyone an ADVC110! I have no ulterior motive, other than to find out once and for all whether there's a little extra visible quality possible with lossless capturing from a VHS source.

[sanlyn]

How many uses for a VHS->DV capture other than playing it on a PC? And I'm still of the opinion that lossy encoding once is better than lossy encoding twice.

[jagabo]

How many uses for a VHS->DV capture other than playing it on a PC?

As an acquisition format if all you have is a DV capture device.

And I'm still of the opinion that lossy encoding once is better than lossy encoding twice.

Of course. The question is how much loss is DV acquisition costing you when it comes to digitizing VHS.

[jagabo]

There's also the issue of chroma placement:

http://www.mir.com/DMG/chroma.html

This effects how colors are interpolated when downscaling (4:4:4 to 4:1:1 or 4:2:0) or upscaling (4:1:1 or 4:2:0 to 4:4:4) chroma.

Here's a modified version of 2bdecided's script to build his test pattern, with a few more options. It requires AviSynth 2.6 though:

y=ImageSource("y.tif",start=0,end=2500,fps=25).ConvertToYV24(matrix="PC.601")
u=ImageSource("u.tif",start=0,end=2500,fps=25).ConvertToYV24(matrix="PC.601")
v=ImageSource("v.tif",start=0,end=2500,fps=25).ConvertToYV24(matrix="PC.601")

YtoUV(u,v,y)

# stop here to output YUV 4:4:4 for software that supports it

# optional:
# the source images were constructed to have all transitions on even boundaries
# the following line forces every other frame to have transitions on odd boundaries
Interleave(last.Subtitle("even"), last.Crop(1,1,-0,-0).AddBorders(0,0,1,1).Subtitle("odd"))

# MPEG2 is the default chroma placement for ConvertToYV12
# you can also specify DV or MPEG1
ConvertToYV12(ChromaOutPlacement="MPEG2")

[Brad]

Ever since I learned they were different, I've been super confused by how interlaced 4:2:0 chroma with MPEG-2 siting compares to PAL DV siting and how upsampling while assuming the opposing siting affects each.

I mean, no one re-sites their PAL DV's chroma when they convert to DVD, do they? Most people are told they're both 4:2:0 and continue on their merry way.

The additional vertical blur is partly due to the PAL filtering, and (I assume) partly due to the DV chroma sub-sampling. Remember that, while you're probably examining this on a progressive PC display, this was sent as interlaced video, so both the PAL filter and the DV subsampling stretch over 3 lines.

How come your M2V uses progressive chroma?

[2Bdecided]

I mean, no one re-sites their PAL DV's chroma when they convert to DVD, do they? Most people are told they're both 4:2:0 and continue on their merry way.

Cedocida automatically converts correctly if you ask it for MPEG-2 4:2:0 chroma. It's got its own control panel.

If you ask for the DV chroma and then pretend it's MPEG-2, you get a 0.5 and 1.5 line chroma shift on u and v (I forget which is which). It's hard to notice the difference, but it is visible if you A/B compare right vs wrong.

How come your M2V uses progressive chroma?

Because that's how I encoded it, rightly or wrongly. The content itself can't be interlaced, because it doesn't move. The SD output of a DVD player can only be interlaced - that's what I meant by it being "sent as interlaced video" - over the analogue hop.

Cheers,
David.

[Brad]

I can capture this test pattern in lossless if that would show anything, but the only player I have that outputs PAL doesn't have an S-Video out (it's a region-switching BD player).

Do you have an NTSC VCR, or just PAL60? (This question is assuming you are even interested in the results of an NTSC DV comparison.)

[sanlyn]

Oops. Wrong thread.

[Brad]

The ADVC110 happily grabbed the full range, and pretty accurately too it seems (though the DVD player I used is an unknown quantity here).

The luma accuracy is impressive, but the missing highly-saturated three corners on the vectorscope is curious.

[jagabo]

The luma accuracy is impressive, but the missing highly-saturated three corners on the vectorscope is curious.

That's because of the blurring of the thin (8 pixel thick) horizontal chroma lines. The blurring is caused by progressive chroma being converted to interlaced chroma at the analog stage.

[Brad]

So it's unrelated to the homogenized colors along the edges of the square?

[jagabo]

So it's unrelated to the homogenized colors along the edges of the square?

Those may contribute a little since they're blurred with the grey background. But it's predominantly the vertical burring of the thin horizontal chroma lines.

[2Bdecided]

That is a fantastic analysis vaporeon800. I think I could re-create it, but can you post the script please?


Just to be clear, this was DVD player out to ADVC110 in. No VHS involved. I posted it because I already had it, and it shows that the ADVC110 get the levels right if they're right on the input. It does seem to slightly clip the very extremes of the chroma, but I don't know if u=255 or v=255 can even go through an s-video Y/C signal chain, so it may not be the ADVC110 at fault. I don't know if anything ever has u or v outside the 16-240 range anyway - it's only Y that I've seen outside 16-235.


Kind of off topic, but talking about illegal colours: I've found that HDMI uses black=16 white=235 (i.e. not the expanded "PC" 0-255 range, but the normal YUV range even for RGB), and some current TVs with certain picture modes correctly display the 235-255 part of the range (i.e. the super whites). Anyone know of any PC media players that will do the same? I guess I could try changing the contrast control in the ones I have.


(Most HDMI devices send/receive YUV by default, but will fall back to RGB when one of them can't handle YUV).




Back on topic: I've tried some really nasty examples, and I can't find one where going through DV makes any visible difference to the final result. I thought I'd found some blocking, but it turned out to be due to the denoiser and was present even when using the lossless source. Subtracting the final result (from the lossless source) from the final result (from the DV source) doesn't reveal any blocking at all - just a very tiny shift to some colours. If I can stop the denoising from blocking I'll post it as an example.




Does anyone remember any threads where there's a lossless (Lagarith, HuffYUV, etc) VHS source, and an AVIsynth script posted that improves it? I thought there'd be hundreds, but the ones I've found either never get anywhere, or use AVIsynth and Neat Video and VirtualDUB - it's rather hard to run all three on two different files and be 100% sure you've got both sets of settings identical. Just AVIsynth would be my preference. Of the ones that meet that criteria, the VHS is so bad that I think people will rightly complain that the results either way are shockingly bad so how could anyone ever expect to see a difference. I thought maybe a really bad source would upset DV, but it hasn't.


Cheers,
David.

[jagabo]

All the HDMI devices I have allow you to specify whether YUV and RGB are 0-255 or 16-235 -- usually with a "high/low" option or some such. And for 16-235 input all the TVs I have allow you to adjust the brightness contrast controls to so as not to crush super blacks and super brights. Most of the software media players in Windows allow that too (either in the software or via the display driver applet in Control Panel).

Oh regarding Vaporeon800's images: He used VideoScope for the luma graph and UV plot at the bottom of the main image, Histogram to the right. UtoY and VtoY were used to isolate the U and V planes as Y, followed by VideoScope for the histograms of those planes below them. I don't know if he cuts and pastes all the images in an editor or does it all in a script.

[Brad]

It's a hilariously big script. I imagine you might be able to simplify it.

# AlignExplode v1.0
#
#   Requires VideoScope plugin. YUY2 input only!
#
#   Parameters are a text Label and Left, Right, Top, Bottom padding. L/R must be even (mod2).
#
function AlignExplode(clip c, string Label, int L, int R, int T, int B)
{
    c
    # Temp 
    TempChromaU = UtoY().VideoScope("bottom",false,HistoTypeBottom="Y")
    TempChromaV = VtoY().VideoScope("bottom",false,HistoTypeBottom="Y")
    TempYCScope = VideoScope("both",false,"Y","Y","UV").Crop(0,height,-0,-0)

    # Layer 1
    Pane1Source = ConvertToRGB().AddBorders(L,T,R,B,color_white)
    Pane2Hist   = Histogram().Crop(width,0,-0,-0).ConvertToRGB().AddBorders(0,T,0,B,color_white)
    Pane3Chroma = TempChromaU.Crop(0,0,width/2,height).ConvertToRGB().AddBorders(L/2,T,R/2,B,color_white)
    Pane4Chroma = TempChromaV.Crop(0,0,width/2,height).ConvertToRGB().AddBorders(L/2,T,R/2,B,color_white)
    Layer1      = StackHorizontal(Pane1Source,Pane2Hist,Pane3Chroma,Pane4Chroma)

    # Layer 2
    Pane5YScope = TempYCScope.Crop(0,0,width,256).ConvertToRGB().AddBorders(L,0,R,0,color_white)
    Pane6CScope = TempYCScope.Crop(width,0,-0,-0).ConvertToRGB()
    Pane7UScope = TempChromaU.Crop(0,height,-0,-0).ConvertToRGB().AddBorders(L/2,0,R/2,0,color_white)
    Pane8VScope = TempChromaV.Crop(0,height,-0,-0).ConvertToRGB().AddBorders(L/2,0,R/2,0,color_white)
    Layer2      = StackHorizontal(Pane5YScope,Pane6CScope,Pane7UScope,Pane8Vscope)

    # Output stack
    StackVertical(Layer1,Layer2)
    AddBorders(0,20,0,0,color_black).Subtitle(Label,size=19,x=2,y=0,font="Courier New").Subtitle("Chroma only",size=19,x=978+L+R,y=0,font="Courier New")

    return last
}