Don't capture interlaced video in 4:2:0

[deleted user]

This is just something I noticed to be a problem with my own captures of analogue sources. For example if you capture an interlaced VHS tape to a DVD, then you are losing important information.

The reason is simple; DVDs always have a color sampling of 4:2:0. As you all know, this means that for every 4 pixels, only one pixel of color information is saved.

Now, the way interlaced material is saved digitally is with alternating lines.

Line 1: Frame 1, Line 1
Line 2 : Frame 2, Line 1
Line 3: Frame 1, Line 2
Line 4: Frame 2, Line 2
Line 5: Frame 1, Line 1
Line 6: Frame 2, Line 2

And so on.

You can probably already see the problem. For every second frame, you are losing the complete color information!

This means that if you deinterlace your video back to 50/60fps, then you have no color information for every second frame. A simple deinterlaced will simply copy the color information of the odd frames to the even frames, thus creating weird looking frames like this:




On the left side the erroneous frame with the color information taken from the previous frame, on the right side the frame as it should actually look. (This example here is made in Photoshop from a normal video, but it pretty accurately represents the artifacts you will get when you do this).

There are some deinterlacers that will temporally interpolate the color channel for this reason afaik, but this won't work properly with very fast movement and lead to ugly artifacts nonetheless. That information is permanently lost and effectively, every second frame is only black & white.

This is actually the case with all HDTV broadcasts, which are all saved in 4:2:0. If you use a deinterlacer and framehop through actual HDTV footage with 50 individual interlaced frames (or fields, if you prefer that term), every second frame/field will have the color information of the former frame.

The solution? When capturing an interlaced source, always capture in the very least in 4:2:2:



With 4:2:2, you still have horizontal chroma subsampling, but vertically, you keep the full resolution. This is necessary to preserve full color information in interlaced material.

If you capture an interlaced source to DVD or a PAL miniDV camera, both of which use 4:2:0 color subsampling, you will lose color information of every second field. This especially applies to TV broadcasts with actual 50 frames/fields and most likely all your old VHS/Hi8/Video8 home videos. So, it's better to capture to 4:2:2 or 4:4:4 and if you want to deliver in 4:2:0, it's best to deinterlace first.

Interestingly, NTSC miniDV uses 4:1:1 subsampling which actually preserves separate lines. Though in this case, your horizontal chroma resolution is quartered, so it's doubtful whether that's worth it.

I believe this is particularly important for Laserdisc capturing, which afaik are often (or always?) interlaced. I think you can probably still get the proper frames back if it's just a 24p stream in a 60i stream (telecine), but if it's actual 60fps content, you are losing all the color of half the frames. Which may apply to any kind of live recordings, interviews, behind the scenes footage, and also some company logos/trailers/intros and such, which are often not produced in 24p.

Thoughts?

[zoidberg]

Pretty common knowledge, I always use LAGS YUY2 for the initial capture. Even PAL material, which has no pulldown can still suffer due to bad field order or video edits causing breaks in cadence. Obviously the old CRTs were more forgiving of this being interlaced displays themselves.

AFAIK DVD recorders also record at 4:2:2, at least mine does. But then you're still dealing with the MPEG2 compression introduced during recording.

[deleted user]

Ah yes, I sometimes tend to spout the obvious, apologies. It's just that I've seen quite some rips of interlaced material in 4:2:0 codecs (for example the normal x264 settings or DVDs), so I thought it would be nice to at least have the information on here, even if it's just basic knowledge.

Pretty sure DVDs can only be 4:2:0 though, unless it's some specific format that's incompatible with the standard:


At least according to wikipedia, the standard requires 4:2:0. Similar to Blu Rays.

I also don't recall seeing a 4:2:2 DVD, but of course that doesn't mean much in itself.

[zoidberg]

Right you are Tom, DVD is 4:2:0. A while back I captured LD to DVD-R and it IVTC'd properly, I took that as meaning it was 4:2:2. Turns out it is 4:2:0, my recorder was just splitting the chroma across the fields properly.

[deleted user]

Yeah I think that it's not actually a problem for telecined content because since each frame basically exists at least twice, one of them is gonna have the correct chroma. It probably really applies more to TV-type content with individual fields.

[althor1138]

Telecined content will also have chroma problems if captured in yv12 colorspace. I seem to remember the x9 project having this problem.

[deleted user]

Telecined content will also have chroma problems if captured in yv12 colorspace. I seem to remember the x9 project having this problem.

Possibly. It would probably depend on the de-telecine method used. For example if you use an automated one that detects similar frames, there's a high chance it would detect repeated frames with different chroma as separate individual frames and mess up the image. On the other hand, if you use a method that exploits the repetitive nature of the telecine (some form of SelectEvery, I think in most cases you will be able to get a decent picture back. I could be wrong though, this is just a guess based on my experience.

[althor1138]

I'm not sure if it was ever verified what the exact problem was but in any case, like you said, always capture at 4:2:2 or higher. I know that they manually ivtc'ed the x9 project though.

Here's the link to where it's discussed:
https://originaltrilogy.com/topic/The-X9...e/5#597977

[deleted user]

I'm not sure if it was ever verified what the exact problem was but in any case, like you said, always capture at 4:2:2 or higher. I know that they manually ivtc'ed the x9 project though.

Here's the link to where it's discussed:
https://originaltrilogy.com/topic/The-X9...e/5#597977

Yep that screenshot and problem description looks like it exhibits exactly this problem. I'm 99.9% confident that capturing or converting to 4:2:0 at some point in the process is the reason for this.

[spoRv]

Other reason could be that X-9 has a digital 3D comb filter, and this *could* be responsible of the smearing - at least, in one of its settings...

Also, all analog consumer video standards were YUV 4:2:2 - VHS, Laserdisc etc. - so no point to capture it 4:4:4; VHS has a so bad chroma resolution that could be indeed captured 4:1:1 @ 352x480/576 without losing neither luma nor chroma info, while laserdisc, even if its luma resolution is lower than DVD and its chroma resolution is lower than half the luma, would lose some colour info if captured 4:1:1 - so, I always recommend to capture laserdisc 4:2:2 @ 720x480/576 - I use HuffYUV as it's fast, but probably Lagarith would be equally good.