DV CODEC for VHS Capture?

[davideck]

There is also the common sense that a 4:1:1 compression from source that usually gets held in 4:2:0 and 4:2:2 is being compressed more than normal, and therefore loss.

There is no compression here.
4:2:2 and 4:1:1 both significantly oversample VHS luma and chroma to the point where no
compression takes place.
A VHS source is not held in 4:2:0 or 4:2:2. Rather, it's luma and chroma are bandwidth
reduced down to something roughly equivalent to 2 : 0.33 : 0.33.

[jagabo]

Ok, what about when transcoded to MPEG which is TFF and using the fourth case above? Wouldn't playback become 3-2-5-4-7-6...?

MPEG can be either TFF or BFF. Just use the right setting.

[lordsmurf]

There is also the common sense that a 4:1:1 compression from source that usually gets held in 4:2:0 and 4:2:2 is being compressed more than normal, and therefore loss.

There is no compression here.
4:2:2 and 4:1:1 both significantly oversample VHS luma and chroma to the point where no
compression takes place.
A VHS source is not held in 4:2:0 or 4:2:2. Rather, it's luma and chroma are bandwidth
reduced down to something roughly equivalent to 2 : 0.33 : 0.33.

The 4:1:1 and 4:2:0 and 4:2:2 ARE THE COMPRESSION. It is colorspace compression. What I'm saying is that you almost always need 4:2:0 and 4:2:2 to retain all the data found in VHS.

Theory of numbers does not work here, be it a simple issue of more complex equations being required, or that DV theory does not live up to practicality, or both. I honestly think the equations are too simplistic, not accounting for the various issues of VHS, and then DV codecs do not work as well as they are theorized to work.

So 4:1:1 is compressed more than normal ("normal" being 4:2:0 and 4:2:2), and VHS does not look good in this DV colorspace compression.

It's not uncommon for VHS -> MPEG off a DVD recorder to have better color/contrast value (transparent to source) than a DV capture software-converted to MPEG (not realtime).

[enbidia]

Ok, what about when transcoded to MPEG which is TFF and using the fourth case above? Wouldn't playback become 3-2-5-4-7-6...?

MPEG can be either TFF or BFF. Just use the right setting.

Is there a flag in the MPEG2 file to tell the player which field order to use?

[lordsmurf]

Ok, what about when transcoded to MPEG which is TFF and using the fourth case above? Wouldn't playback become 3-2-5-4-7-6...?

MPEG can be either TFF or BFF. Just use the right setting.

Is there a flag in the MPEG2 file to tell the player which field order to use?

Yes. Use GSPOT 2.52 to read the file, and it will show PROG or TFF or BFF for field order (or lack thereof, progressive).

[enbidia]

thanks, did use it to read a DVD MPEG2 file but still couldn't locate where the field order is being indicated.

[jagabo]

[davideck]

There is also the common sense that a 4:1:1 compression from source that usually gets held in 4:2:0 and 4:2:2 is being compressed more than normal, and therefore loss.

There is no compression here.
4:2:2 and 4:1:1 both significantly oversample VHS luma and chroma to the point where no compression takes place.
A VHS source is not held in 4:2:0 or 4:2:2. Rather, it's luma and chroma are bandwidth
reduced down to something roughly equivalent to 2 : 0.33 : 0.33.

The 4:1:1 and 4:2:0 and 4:2:2 ARE THE COMPRESSION. It is colorspace compression. What I'm saying is that you almost always need 4:2:0 and 4:2:2 to retain all the data found in VHS.

The compression occurred when the VHS recording was made.
The bandwidth of the resulting VHS playback signal requires a minimum sampling of about 2 : 0.33 : 0.33. There is no additional compression introduced by oversampling this requirement up at 4:1:1.

Sampling VHS at 4:2:0 does introduce additional compression by halving the chroma vertical resolution.

I am not disputing any DV codec issues you may have, but I am suggesting that they are not due to sampling the VHS bandwidths at 4:1:1.

[enbidia]

Mine is truly missing.

[lordsmurf]

I am not disputing any DV codec issues you may have, but I am suggesting that they are not due to sampling the VHS bandwidths at 4:1:1.

I still think it is a part of the problem. To what degree, I don't know.

[guns1inger]

To Enbidia - use 2.52b beta instead

[enbidia]

To Enbidia - use 2.52b beta instead

thanks, its working now

[enbidia]

There is also the common sense that a 4:1:1 compression from source that usually gets held in 4:2:0 and 4:2:2 is being compressed more than normal, and therefore loss.

There is no compression here.
4:2:2 and 4:1:1 both significantly oversample VHS luma and chroma to the point where no compression takes place.
A VHS source is not held in 4:2:0 or 4:2:2. Rather, it's luma and chroma are bandwidth
reduced down to something roughly equivalent to 2 : 0.33 : 0.33.

The 4:1:1 and 4:2:0 and 4:2:2 ARE THE COMPRESSION. It is colorspace compression. What I'm saying is that you almost always need 4:2:0 and 4:2:2 to retain all the data found in VHS.

The compression occurred when the VHS recording was made.
The bandwidth of the resulting VHS playback signal requires a minimum sampling of about 2 : 0.33 : 0.33. There is no additional compression introduced by oversampling this requirement up at 4:1:1.

Sampling VHS at 4:2:0 does introduce additional compression by halving the chroma vertical resolution.

I am not disputing any DV codec issues you may have, but I am suggesting that they are not due to sampling the VHS bandwidths at 4:1:1.

I would rather have it sampled at 4:2:2 at the very least just for sure. Below is the explanation:


We might think of de-facto NTSC as 4:0.5:0.5 since it takes the equivalent of eight luminance pixels (broadcast standard and limit) to transition from one color to another. The pixel analogy is not quite correct. The color transition in analog video could span "pixels 4 through 11" as opposed to "pixels 1 through 8" or "pixels 9 through 16" while in digital video (4:1:1) color is shared by pixels 1 through 4 and pixels five through eight, and never pixels three through six. We would expect that digitizing NTSC analog video as 4:0.5:0.5 would produce noticeable degradation, using 4:1:1 or 4:2:0 would produce some degradation, and using 4:2:2 would produce negligible degradation.

http://members.aol.com/ajaynejr/vidcol2.htm#Broad2


IMO use lossless compression if possible. I think DCT and quantization used in all DV formats would further destroy spatial or luminance details of the video.

[davideck]

What I'm saying is that you almost always need 4:2:0 and 4:2:2 to retain all the data found in VHS.

You often suggest that sampling VHS at 352x480 is sufficient for DVD capture. 352 sampling is roughly equivalent to 2:1:0 ?

We would expect that digitizing NTSC analog video as 4:0.5:0.5 would produce noticeable degradation, using 4:1:1 or 4:2:0 would produce some degradation, and using 4:2:2 would produce negligible degradation.

I do not disagree with this statement, but I would note that the VHS luma/chroma bandwidths are about half that of the NTSC analog video standard.

[lordsmurf]

. 352 sampling is roughly equivalent to 2:1:0 ?

No. Separate concepts.

[davideck]

. 352 sampling is roughly equivalent to 2:1:0 ?

No. Separate concepts.

How so?

[enbidia]

We would expect that digitizing NTSC analog video as 4:0.5:0.5 would produce noticeable degradation, using 4:1:1 or 4:2:0 would produce some degradation, and using 4:2:2 would produce negligible degradation.

I do not disagree with this statement, but I would note that the VHS luma/chroma bandwidths are about half that of the NTSC analog video standard.

Nevertheless color degradation theoretically are still there however at a lesser degree. It depends also on how big and sharp the display which may or may not be noticeable. In theory if you have alternating colored bars of 30 lines thick -- VHS chroma resolution -- and start from a random position and resampled it at 4:1:0 or 4:2:2 there will be slight loses to the color values in some portions of those bars as well as some softness at the edges which vary in degree depending on the sensitivity of the eye to that particalar color and colors beside.

[davideck]

enbidia -

If you are suggesting that a 600 Khz bandwidth chroma signal cannot be sampled at 3.375Mhz (4:1:1) and reconstructed without noticeable degradation, then I disagree.

I would also note that at 4:1:1, the .6 Mhz VHS chroma bandwidth is sampled at a 5.6X ratio while the 3 Mhz VHS luminance bandwidth is only sampled at a 4.5X ratio. The higher the ratio, the better.

For broadcast video quality, 4:2:2 sampling is used to sample a 6 Mhz luminance bandwidth at a 2.25X ratio and a 1.5 Mhz chroma bandwidth at a 4.5X ratio.

[edDV]

We would expect that digitizing NTSC analog video as 4:0.5:0.5 would produce noticeable degradation, using 4:1:1 or 4:2:0 would produce some degradation, and using 4:2:2 would produce negligible degradation.

I do not disagree with this statement, but I would note that the VHS luma/chroma bandwidths are about half that of the NTSC analog video standard.

Nevertheless color degradation theoretically are still there however at a lesser degree. It depends also on how big and sharp the display which may or may not be noticeable. In theory if you have alternating colored bars of 30 lines thick -- VHS chroma resolution -- and start from a random position and resampled it at 4:1:0 or 4:2:2 there will be slight loses to the color values in some portions of those bars as well as some softness at the edges which vary in degree depending on the sensitivity of the eye to that particalar color and colors beside.

Big difference between low noise studio NTSC (decoded to YUV by a quality comb filter decoder) and the crosscolor contanimated 500KHz UV that results from the VHS color under recording process. Studio quality decoding would produce 1-1.5MHz of true chroma bandwidth that may benefit from 6.75MHz (4:2:2) oversampling vs 3.38MHz (4:1:1).

It could be argued that low pass filtering VHS UV to no more than 500KHz may improve the capture by minimizing effects of luminance crosscolor contamination produced by sampling a wider bandwidth. All of this would require experimentation using a variety of video samples.

Theory says that oversampling a clean signal can reduce the demands (sharpness) of the analog Nyquist low pass filter (before A/D) but if the signal is very noisy in the higher frequencies, logic would say a tighter low pass filter before A/D would be the best strategy for VHS rather than increasing oversampling and relaxing the Nyquist filter. Unfortunately this makes for a very expensive capture card since these filters cost >$75 each for Y U and V.

VHS luminance should be sharply filtered above 3MHz before A/D since the VHS FM recording process prohibits any true luminance signal to exist above 3MHz. All that lives up there is nasty noise from chroma cross color contamination. Oversampling VHS luma just doesn't make sense, especially when using a cheap Nyquist filter.

Contrary to conventional logic, VHS capture needs a more expensive capture card design than does relatively cleaner analog Betacam.

[edDV]

S-VHS has the same issues for chrominance but different issues for luminance Y.

In the case of S-VHS, luminance bandwidth exists out to 4.5 MHz so 13.5MHz oversampling may help (Nyquist minimum 9MHz). The problem with S-VHS is cross color contamination in the 3.58 +/- 0.5MHz frequencies if the original recorded signal was composite. Selective bandpass filtering around 3.58MHz before A/D may improve the picture.

If the recorded signal was pure Y/C component (e.g. a camcorder), then S-VHS luminance can be fairly clean.

[jagabo]

This is what happens when you capture a low bandwidth source with DV:



I think you can see why many people don't like DV. Over the years I've seen many people here post samples from Canopus ADVC devices with this type of problem. Here's an example:

https://forum.videohelp.com/viewtopic.php?t=261543

[enbidia]

enbidia -

If you are suggesting that a 600 Khz bandwidth chroma signal cannot be sampled at 3.375Mhz (4:1:1) and reconstructed without noticeable degradation, then I disagree.

I would also note that at 4:1:1, the .6 Mhz VHS chroma bandwidth is sampled at a 5.6X ratio while the 3 Mhz VHS luminance bandwidth is only sampled at a 4.5X ratio. The higher the ratio, the better.

For broadcast video quality, 4:2:2 sampling is used to sample a 6 Mhz luminance bandwidth at a 2.25X ratio and a 1.5 Mhz chroma bandwidth at a 4.5X ratio.

Hi, where did you get the 6 MHz luminance bandwdith? Broadcast NTSC TV has only around 4.2 MHz for the luminance and 0.5 MHz typically for the chroma. The article http://members.aol.com/ajaynejr/vidcol2.htm#Transition refers 4:0.5:0.5 as its digital equivalent where 4 is 4.2 Mhz and 0.5 is 0.5 MHz which the color resolution is roughly equivalent to VHS but recommended 4:2:2 oversampling which the 4 is 13.5 MHz and the 2 is 6.75 MHz

[enbidia]

We would expect that digitizing NTSC analog video as 4:0.5:0.5 would produce noticeable degradation, using 4:1:1 or 4:2:0 would produce some degradation, and using 4:2:2 would produce negligible degradation.

I do not disagree with this statement, but I would note that the VHS luma/chroma bandwidths are about half that of the NTSC analog video standard.

Nevertheless color degradation theoretically are still there however at a lesser degree. It depends also on how big and sharp the display which may or may not be noticeable. In theory if you have alternating colored bars of 30 lines thick -- VHS chroma resolution -- and start from a random position and resampled it at 4:1:0 or 4:2:2 there will be slight loses to the color values in some portions of those bars as well as some softness at the edges which vary in degree depending on the sensitivity of the eye to that particalar color and colors beside.

Big difference between low noise studio NTSC (decoded to YUV by a quality comb filter decoder) and the crosscolor contanimated 500KHz UV that results from the VHS color under recording process. Studio quality decoding would produce 1-1.5MHz of true chroma bandwidth that may benefit from 6.75MHz (4:2:2) oversampling vs 3.38MHz (4:1:1).

It could be argued that low pass filtering VHS UV to no more than 500KHz may improve the capture by minimizing effects of luminance crosscolor contamination produced by sampling a wider bandwidth. All of this would require experimentation using a variety of video samples.

Theory says that oversampling a clean signal can reduce the demands (sharpness) of the analog Nyquist low pass filter (before A/D) but if the signal is very noisy in the higher frequencies, logic would say a tighter low pass filter before A/D would be the best strategy for VHS rather than increasing oversampling and relaxing the Nyquist filter. Unfortunately this makes for a very expensive capture card since these filters cost >$75 each for Y U and V.

VHS luminance should be sharply filtered above 3MHz before A/D since the VHS FM recording process prohibits any true luminance signal to exist above 3MHz. All that lives up there is nasty noise from chroma cross color contamination. Oversampling VHS luma just doesn't make sense, especially when using a cheap Nyquist filter.

Contrary to conventional logic, VHS capture needs a more expensive capture card design than does relatively cleaner analog Betacam.

IMO I think analog noises are more subtle and forgiving to the eye than digital artifacts. Analog devices imperfections seem to be more gentle and gradual and fade in subpixel smoothness from its area of origin. Digital imperfections are more annoying. It appears in pixels blocks and do not fade away gradually but suddenly which to the eye means sharp and noticeable.

[davideck]

Hi, where did you get the 6 MHz luminance bandwdith? Broadcast NTSC TV has only around 4.2 MHz for the luminance and 0.5 MHz typically for the chroma.

I agree that my numbers were off. 4.2 Mhz for luminance is a better number. At 4:2:2, that's a sampling ratio of 3.2X.

The Q color channel is limited to 0.5 Mhz, but the I color channel has a bandwidth of 1.3 Mhz. At 4:2:2, that's a ratio of 5.2X

http://www.sencore.com/newsletter/Jan02/JanNews/Understanding%20and%20Measuring%20Part%20I.htm

[edDV]

This is what happens when you capture a low bandwidth source with DV:



I think you can see why many people don't like DV. Over the years I've seen many people here post samples from Canopus ADVC devices with this type of problem. Here's an example:

https://forum.videohelp.com/viewtopic.php?t=261543

So you are doing a saturated barber pole documentary to Digital Betacam 4:2:2 from a wideband source? Most people would consider that a poor way to spend your bitrate budget. You normally want to preserve luminance detail so you compromise chroma bandwidth to the medium available. DVD isn't that bad at 4:2:0 (roughtly the same as 4:1:1) and I know of nobody recording to D1 4:2:2 or 10bit luminance at home.

But you said your source was VHS. At 500 KHz color behind the luminance is only blobs with about 30-50 blob changes over a line.

Explain to us what you are trying to do again?

BTW here is an off cable DV 4:1:1 capture over S-Video (Canopus ADVC-100). Not bad if you ask me but you are claiming 4:1:1 would perform worse if the source was low res VHS? Worse than the source?



A good read on the subject here.
http://www.lafcpug.org/Tutorials/basic_chroma_sample.html

[edDV]

That is an odd test pattern you have there. What is it supposed to measure?
Looks like low luminance variation combined with large chroma anplitude + phase variation.

[enbidia]

Hi, I can clearly see the chroma resolution inferiority and artifacts of DVCAM 4:1:1 vs uncompressed 4:2:2. Despite passing thru low resolution analog BetacamSP and then redigitized the chroma is still clearly superb to the higher resolution DV 4:1:1 direct.





Pls. pay attention to the shirt. The creases are gone.


and his comment about the chroma:

The chroma on the BetacamSP is very interesting as it is significantly softer than the 4:2:2 chroma of the Digital Betacam, but also appears slightly sharper than the 4:1:1 chroma of the DVCAM. The BetacamSP to DVCAM dubs look worst of all, which is nor surprising given that they have gone down an extra generation compared to the other examples.

I don't think it is slightly sharper but much sharper. If you count the creases on the front part of the shirt there are at least seven there. For 4:2:2 uncompressed despite redigitzed from Betacam SP most details are still intact but for DVCAM the looks are very different.

[jagabo]

That is an odd test pattern you have there. What is it supposed to measure?
Looks like low luminance variation combined with large chroma anplitude + phase variation.

Exactly. If you count the lines you'll see there are about 40 across the 720 pixel wide frame. VHS will have little problem with this (you may lose a little saturation).

As I think you are aware, when most consumer grade DV decoders decode from 4:1:1 back to 4:4:4 they simply replicate the chroma values four times, or for 4:1:1 to 4:2:2 they duplicate chroma samples. This leads to the obvious stripe/block artifacts that many people complain about (on this type of material). I used ffdshow's DV decoder in my sample image.

So even though DV's 4:1:1 chroma subsampling is theoretically sufficient for a VHS source, real-world implementations can lead to obvious artifacts.

[enbidia]

I think the best way to experiment with these various dv codecs is to generate various smallest barely readable text still and moving at all screen positions using just its chroma channel on VHS, SVHS, Betacam etc. Then swap the chroma channel values to the luminance channel during capture and encoding. If ever these text are unreadable then we may know the oversampling rate of this codec isn't sufficient for that format.

That is an odd test pattern you have there. What is it supposed to measure?
Looks like low luminance variation combined with large chroma anplitude + phase variation.

Exactly. If you count the lines you'll see there are about 40 across the 720 pixel wide frame. VHS will have little problem with this (you may lose a little saturation).

As I think you are aware, when most consumer grade DV decoders decode from 4:1:1 back to 4:4:4 they simply replicate the chroma values four times, or for 4:1:1 to 4:2:2 they duplicate chroma samples. This leads to the obvious stripe/block artifacts that many people complain about (on this type of material). I used ffdshow's DV decoder in my sample image.

So even though DV's 4:1:1 chroma subsampling is theoretically sufficient for a VHS source, real-world implementations can lead to obvious artifacts.

[edDV]

Hi, I can clearly see the chroma resolution inferiority and artifacts of DVCAM 4:1:1 vs uncompressed 4:2:2.

Well duhhh!
We would all like $40,000 Digital Betacam camcorder acquisition performance vs. $3000 DVCAM. No brainer there.

The jist of that article was to show 4:1:1 material can be better integrated into an uncompressed 4:2:2 production by interpolating the missing chroma pixels rather than simply repeating the existing pixel.

The different sampling resolutions optimize differently when all the economic factors of the production are taken into account. Uncompressed capture to 4:2:2, processing in 4:2:2 and then encoding can get great results but it can be very costly in time and computer resource.

Using DV as an intermediate format has productivity advantages if you already own a pass through DV camcorder or hardware DV transcoder. Editing and filtering are simple and computer resources are more efficiently used.

Tradeoffs are few but they are there and the effect shown above tends to be more noticed where luma changes are small and chroma levels abnormally high. Normally the higher resolution luminance masks the rougher chroma edges. I was looking for some examples in this DV PVR capture last night. Normal luminance contrast seems to produce acceptable edges even when chroma saturation is high.



But when luminace contrast was low and chroma saturation high, those edge effects appeared. If this were VHS, the noise would be extremely high for low reds.



The pictures above were from the encoded MPeg2
(i.e. S-Video->ADVC-100 (DV)->Vegas DV timeline->MPeg2 @7,000Kb/s VBR)

This one shows extreme luminace contrast and the limits of 8 bits. But nothing is clipping.


PS: Before some noob complains about compression artifacts in these images, notice that we are talking about edge smoothness. Each uncompressed frame of 720x480 video is as follows:

4:1:1 = 720x480x1.5 Bytes = 518KB
4:2:2 = 720x480x2.0 Bytes = 691KB
4:4:4 = 720x480x3.0 Bytes =~ 1MB

The pictures above are ~ 50KB or ~10x compressed for display here.

DV compresses ~5x so the above DV images are ~2x compressed vs. DV format.