DV vs. lossless differences in colorspace?

[Aya_Rei]

To make a long story short, got a video8 tape from a family member that was crinkled and sent it in to my local transfer service to get it repaired along with digitizing the contents of the tape since I wasn't going to risk any of my equipment on such a tape.

Got the tape back and found out the footage was encoded through DV, thankfully since the tape itself was fixed up (minus the first 40 seconds of the 1 hour recording which is unwatchable) I was able to digitize the rest of it myself.

This thread isn't meant to shame the service who did the tape for me, I'd certainly recommend them for physically damaged tapes, photos, audio, professional tape formats, etc. But instead, I thought it'd be a nice way to show yet another example of the differences between DV's 4:1:1 chroma depth and the 4:2:2 chroma depth of loseless analog.

For the analog captures the Pinnacle 510's proc amp was adjusted to prevent clipping (Brightness was increased by 10, Contrast was decreased by 18) The frame TBC's proc amp was left at it's default values.

Comparisons are below, the last two comparisons were deinterlaced due to the fast motion

https://imgsli.com/MzAzMDY5
https://imgsli.com/MzAzMDcw
https://imgsli.com/MzAzMDcx

[latreche34]

Chroma in the lossless sample is oversaturated, Luma is a little hotter, To be honest the DV sample has the proper levels.

[Aya_Rei]

Oh really? Well I do apologize for messing up, don't really adjust the proc amp values on the Cypress CDM 640 and just leave it at default.

Here's another comparison, this time I decreased the color proc amp control by 4. Everything else (including the brightness and contrast changes on the Pinnacle was left unchanged)

https://imgsli.com/MzAzMDgx

To me anyway, the version with the proc amp decreased by 4 matchs up well with the DV sample in terms of chroma. So I suppose for future reference I'll decrease the color proc amp by 3 from it's default value or so.

Edit - Did a bit more testing, comparing the colors from using the CDM 640 with the color decreased by 3 and without using the CDM 640 at all. To me, it looks like decreasing the color proc amp by 3 is enough to make both results basically the same in terms of chroma.

https://imgsli.com/MzAzMDg2

My apologies, guess this isn't so much of a difference between 4:1:1 and 4:2:2 and moreso a different between using the Cypress CDM 640 Frame TBC at the default color value version decreasing it by 3 in order to make the chroma not oversaturated.

[aramkolt]

Seems in the comparisons that the full frame isn't quite captured with DV? Seems like it is missing the right-most edge at least. Just curious why that might be or if it can be fixed at the time of capture?

Also, what is the preferred method to check for correct color levels after a capture in terms of which of the two samples is most appropriate?

[lordsmurf]

The DV color loss here is substantial. Wow. That's bad.

While the lossless does have boosted saturation (and 99%+ likely caused by the home-quality consumer camcorder), it can be corrected. Better than now, better than the DV.

If you drag these into Photoshop, and boost colors on the DV, you'll notice that the DV is not just reducing the color, but augmenting tints.

The ability to correct color should be your option, not forced by inferior DV conversion.

[aramkolt]

DV color has always been an interesting topic to me. Doesn't seem like there should be anything inherently bad about how DV chips store analog colors as digital equivalents and I wonder if certain DV conversion chips did a better job than others with color accuracy or perceived blockiness. I haven't seen too many direct comparisons of different hardware DV converters to show differences in the results. I might have to do something like that as part of my testing.

I suspect if you were to use something like the AJA Kona LHi and save the video stream as DV within AJA's capture utility that the color value assignments would be indistinguishable from other formats that you could save the stream in instead given that the computer is assigning chroma values for everything (and not a hardware chip).

4:1:1 does still sample every 4th horizontal pixel for color and records data for each horizontal line whereas 4:2:0 throws away every other horizontal line's data. Given the severely low chroma resolution of only 40 per line for VHS, we are still sampling that over 4x more than how it is stored if only 180 samples are stored from each 720 pixel line. VHS does contain unique data for each horizontal line, so throwing away every other line seems worse than sampling an already very low resolution horizontal line less often than 4:2:0 does.

Common_chroma_subsampling_ratios_YCbCr_CORRECTED.svg.jpg

[Gary34]

There’re plenty of threads over converting VHS using DV if people just want to read other people’s mistakes and avoid repeating them.
https://www.digitalfaq.com/forum/myt...t-convert.html

[latreche34]

Quote:

Originally Posted by aramkolt

4:1:1 does still sample every 4th horizontal pixel for color and records data for each horizontal line whereas 4:2:0 throws away every other horizontal line's data. Given the severely low chroma resolution of only 40 per line for VHS, we are still sampling that over 4x more than how it is stored if only 180 samples are stored from each 720 pixel line. VHS does contain unique data for each horizontal line, so throwing away every other line seems worse than sampling an already very low resolution horizontal line less often than 4:2:0 does.

I've argued this over the years but no one seems to agree. In NTSC, DV doesn't give too much emphasis to horizontal chroma resolution since it is low to begin with (around 40 samples for consumer analog tape formats), so Sony though double sampling it will be more than enough, PAL in the other hand throws away an entire field of chroma info and just slaps the adjacent field chroma to it (in other words, every even scan line is given the odd's scan line chroma info).

In the interlaced world of CRT this would not have much effect since our brains can be tricked by temporal events, But in the digital age where the files are de-interlaced by software or by modern displays and refresh rates are doubled, the lack of vertical chroma in PAL will have an effect on the overall chroma quality, sometimes seen as thick horizontal gradual lines of chroma.

[vwestlife]

It dates back long before DV. In the early 1990s, standard-definition digital video in the U.S. was officially defined as 720 luma transitions and 360 chroma transitions per scanline. So right from the start, the chroma resolution was halved, because that's all that a typical CRT could resolve back then.

And I find the chroma pixellation of 4:2:0 (especially on anything bright red) to be far worse than with 4:1:1.

[lordsmurf]

To add...

There is no "4:2:0" or "4:1:1" in the analog realm. That's a digital measurement. 4:1:1 / 4:2:0 is definitely degraded, which is why all broadcasting is 4:2:2. Technically, 4:2:2 is "overkill", but everything else is "underkill" (more lossy). You want to minimize loss, not amplify it. DVD-Video used 4:2:0 for multiple reasons, being a highly compressed consumer delivery format.

Part of this is also "math vs. reality". While the numbers insinuate one thing ("4:1:1 is plenty" = BS), actuality is another. The reason? Analog is not precise, there is slop.

Codecs are also universally lousy.

@vwestlife: Both 4:2:0 and 4:1:1 screw up bright colors. Neither is better, each just sucks differently.

[Aya_Rei]

What interesting points, thanks for the insight.

Anyway, realized that I can also just turn down the saturation of the captures I've done at CDM 640's default value to 0.80 in order to more closer match what they'd look like when I turn the color proc amp down by 3 stages. Guess I'll stick with adjusting saturation and whatnot in post production rather than the Cypress?

I do remember hearing that it's best to correct color related controls (saturation, hue) in post production since with VHS and video8, color can vary wildly from scene to scene and due to differences in camera and recording VCR, basically there are a lot of factors here to consider..

[ThumperStrauss]

On the issue of brightness, I have used Premiere Pro to restore some detail from over-bright objects from the DV capture in order to to make it match the brightness as the same capture done with an ATI 9600XT and with the proc amp settings adjusted in VirtualDub beforehand.

So, you should try that with this footage to see if you can fix the overbright areas in the DV version.

[dpalomaki]

Quote:

I do remember hearing that it's best to correct color related controls (saturation, hue) in post production since with VHS and video8, color can vary wildly from scene to scene and due to differences in camera and recording VCR,

To the extent you can match individual cameras before the shoot.
For capture from analog sources color/exposure correct each tape as best you can before capture while still in analog form. But be sure you are using a good proc amp. This ensures the captured digital file best preserves the dynamic range of the original.
Do final tweaking in post (after capture).

Note that once captured any over exposure/clipping type errors may be baked in.

[timtape]

The important thing is not to make the recording worse in the capture stage. So avoid clipping highlights at capture. If they're already somewhat clipped on the tape, don't clip them more at capture.

[Aya_Rei]

Quote:

Originally Posted by timtape

The important thing is not to make the recording worse in the capture stage. So avoid clipping highlights at capture. If they're already somewhat clipped on the tape, don't clip them more at capture.

Yeah that is true, which is why for more recent captures I've started to adjust the brightness and contrast on the Pinnacle (and leave the Cypress TBC's at their defaults) as oppose to just leaving everything at default, in order to prevent clipping for both black and whites

I can tweak the brightness and contrast in post to match what the footage would look like without the tweaks if need be.