Comparison: Video8 captured via SVideo/TBC On; SVideo/TBC Off; DV

[Bwaak]

A short comparison just for kicks.

Source video: 1991 8-mm video, regular Video8, not Hi8. Captured from Sony Digital8 camcorder three different ways:

• Built-in TBC Off, captured via SVideo into Cineform Filmscan 1
• Built-in TBC On, captured via SVideo into Cineform Filmscan 1
• Built-in TBC On, internal analog to DV conversion, captured via Firewire

No recompression. No audio.

Even with TBC turned off, the video looks very stable. I wonder is it a benefit of 8-mm format over VHS, or just a good recording and good condition of the camcorder.

[Skiller]

I don't have Cineform codecs installed. Is there a way someone could losslessly recompress the files into something more commonly used here on the forum, such as HuffYUV, please?

[Cornucopia]

@Skiller, you should be able to do it with ffmpeg, IIRC.


Scott

[Alwyn]

The same H264 parameters for all would have been sufficient for a comparison. Nobody's ever going to view them in Cine or DV.

[Bwaak]

The same H264 parameters for all would have been sufficient for a comparison. Nobody's ever going to view them in Cine or DV.

If you want to see the captures but do not have Cineform, I can deinterlace and re-encode to H.264 for your viewing pleasure. If you don't want to watch them at all, this is fine too. But don't pass off your opinion as everyone's.

[Alwyn]

But don't pass off your opinion as everyone's.

Yessir.

[lordsmurf]

The same H264 parameters for all would have been sufficient for a comparison. Nobody's ever going to view them in Cine or DV.

If you want to see the captures but do not have Cineform, I can deinterlace and re-encode to H.264 for your viewing pleasure. If you don't want to watch them at all, this is fine too. But don't pass off your opinion as everyone's.

Do not deinterlace.

If serious video hobbyists don't want to hassle with non-standard downloads, 0% chance that anybody else will either. I'm not downloading Cineform files.

Samples are never for "viewing pleasure".

Since you have an axe to grind over TBCs, I have my doubts that you'll post truly objective samples. I see far too much cherry-picking (aka BS samples) when it comes to video. Not just biased users, but companies like Topaz and Blackmagic routinely dole out BS. I still remember TBC ads from the 90s and early 00s, total nonsense claims of abilities.

[Bwaak]

Come on, Alwyn, let's bury the hatchet If you feel I've been rude, I apologize.

Here are the same clips, deinterlaced with MSU Deinterlacer in VDub2 and compressed with H.264, but not upscaled.

[Sharc]

All are pretty stable re. horizontal wiggling.
Footnote: The DV variant has elevated blacks and superwhites (even slightly clipped), loosing details in the bright areas.

[oln]

How stable it is without TBC on the camcorder disabled will depend a lot on what the rest of the capture setup is, and also the footage. I've not found 8mm to be particularly more stable than VHS though a plus is that you do have a set of playback devices, late model sony camcorders, that have a very good and reliable TBC unlike with VHS where the built in stuff is rare/expensive and can be a bit hit and miss.

[lordsmurf]

I wonder is it a benefit of 8-mm format over VHS, or just a good recording and good condition of the camcorder.

This is part of it. Sony 8mm tapes have less trassport "slop" to creates line timing errors. It's not invisible, but not awful. Still best to have line TBC. Some tapes are awful, and it is very tape based.

The real problem with Video8 and Hi8 is excessive dropped frames. Most Digital8 cameras compress to DV before output (A>D>A, not passthrough), so it really obscures tests. It bakes in frame drop issues.

So your test is good, but don't try to make it mean more than it really does.

[Bwaak]

All are pretty stable re. horizontal wiggling.
Footnote: The DV variant has elevated blacks and superwhites (even slightly clipped), loosing details in the bright areas.

Hi Sharc, can you share the script for the levels diagram? I bet you've posted it many times before.

[Sharc]

All are pretty stable re. horizontal wiggling.
Footnote: The DV variant has elevated blacks and superwhites (even slightly clipped), loosing details in the bright areas.

Hi Sharc, can you share the script for the levels diagram? I bet you've posted it many times before.

Here we go:

ffms2("render (video8-to-DV-internal-tbc-on-H264-excerpt-no-audio-60p).mp4",cache=false)  #or your source filter
histogram("classic")

I would suggest to adjust the levels in post with something like

smoothlevels(16,1.0,255,0,235)

which will bring the details in the bright areas back

[lollo]

ffms2("render (video8-to-DV-internal-tbc-on-H264-excerpt-no-audio-60p).mp4",cache=false)  #or your source filter
histogram("classic")

A cropping of the black borders (if any, I did not check the source) is missing

[Sharc]

ffms2("render (video8-to-DV-internal-tbc-on-H264-excerpt-no-audio-60p).mp4",cache=false)  #or your source filter
histogram("classic")

A cropping of the black borders (if any, I did not check the source) is missing

I usually do crop, but it is not mandatory to crop narrow borders (few pixels) for the waveform monitor. They would just be included in the waveform as a faint line.
In the statistical histogram however (histogram("levels")) the black borders will bias the distribution, so yes, in that case black borders should definitely be cropped.

[Bwaak]

I usually do crop, but it is not mandatory to crop narrow borders (few pixels) for the waveform monitor. They would just be included in the waveform as a faint line. In the statistical histogram however (histogram("levels")) the black borders will bias the distribution, so yes, in that case black borders should definitely be cropped.

Can you elaborate regarding "the statistical histogram"? The way I usually verify my black level for Youtube is render several versions WITH black borders, upload to Youtube, switch to theater or full-screen mode and check whether the baked-in black borders brighter than the screen background. With all the knowledge about black and white levels, I still cannot dial them in based on specs/numbers/theory alone, so I just make sure that the borders disappear. THEN I crop them and do the final render.

If "the black borders will bias the distribution", then my approach may produce wrong results? How exactly this bias occurs, and in what environment? Is it applicable only to Avisynth filters?

[Sharc]

Just try and see it:

Waveform monitor:

histogram("classic")

Statistical historgram:

histogram("levels")

Also see the description in the avisynth wiki.
http://avisynth.nl/index.php/Histogram

General info about histogram, see:
https://en.wikipedia.org/wiki/Histogram

Bias of the histogram("levels"):
Usually one is interested in the distribution of the luma values (means frequency of occurrence (=histogram) of the luma values) of the active picture.
Black borders have nothing to do with the active picture, but unless they are cropped they will appear in the histogram as a peak at the corresponding luma level. In a normalized view this dominant peak will "push" the rest of the histogram down, hence biasing the histogram. Due to the "black" borders which are represented by a single luma value, this luma value of the borders appears statistically frequently, means more frequent than the remaining luma values of the "natural" active picture).
Maybe a confusing and oversimplified explanation, so better just try it and see yourself, and remember basic statistics from college. Or perhaps a better teacher than me may chime in ......

It is generally good and recommended practice to keep the luma values of 8bit YCbCr (YUV) video in the Y (=luma) range 16 ...... 235 for final distribution in order to avoid unnecessary surprises and trouble regarding color distortions and loss of details/gradients. Fine tuning is usually done in post processing.

Added:
Calibrating the "black level" of a video encode using your TV monitor's background as a reference can be misleading as the result depends on the TV settings (brightness, contrast, gamma).

Added2:
Now coming back to your captures:
- The first picture is the S-video capture with the luma in the range (16<Y<235) which is perfect.
- The second picture is the DV capture with elevated darks and brights. The brights go up to 255, and some structure in the bright ground below the red wagon is lost. The loss happens when the YUV of the DV's superwhites (even slightly clipped) is converted to RGB for viewing. As the clipping is very minor, the 'losses' can mostly be recovered with a levels adjustment (3rd picture).