Hello guest, if you like this forum, why don't you register? https://fanrestore.com/member.php?action=register (December 14, 2021) x


Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
The "joy" of de-interlacing and weird stuff I have noticed...
#1
Those of us that have worked on various projects have, at some point, had to deal with de-interlacing interlaced footage. Now there are numerous different ways of doing this, with some rather decent results to complete nightmares. All of this is usually dependent on the resolution of the video, the bitrate (the lower it is the more macro-blocking/noise/artefacts get introduced) and the Codec used (mostly MPEG2). I don't want to discuss the various methods as everyones "mileage will vary" depending on numerous factors.

However, while working on the "Something Evil" project, that finally has a new source (more in this thread) I noticed something in the de-interlaced footage. (I was using TDeint in Avidemux 2.5 exporting using Huffyuv)

I really like the real-time preview in Sony Vegas of multiple different video source on a frame-by-frame basis that is extremely easy to navigate. This is how I noticed that the footage that I de-interlaced (top-field-first) on many scene/shot change would have a chroma-bleed of frame A into frame B or chroma-bleed of frame B into frame A, at various spots, interchangeable, with no particular pattern. I thought to myself, "what if I reverse the field order when de-interlacing the footage?" The result is now that the chroma-bleeds happen at EXACT opposite frames, therefore, each frame can be salvaged.

Andrea suggested screenshot comparisons and this is a great idea Big Grin

frame 84694 and comparison between de-interlaced top-field-first and bottom-field-first:
http://screenshotcomparison.com/comparison/128834

frame 84695, same comparison as last frame, but you can see the chroma-bleed reversal in relation to choosing different first field:
http://screenshotcomparison.com/comparison/128836

frame 3851 was chosen here for a few reasons... it is NOT at a scene change and shows how more detail is brought out by overlaying the bottom-field-first at 50% over the top-field-first helps produce a slightly sharper image as all horizontal info from both fields in present in the single progressive frame instead of just half (granted, at this lower resolution the difference is not as big, but I suspect this could help a lot with HD sources)... especially noticeable in the "for sale" sign and phone number underneathWink
http://screenshotcomparison.com/comparison/128837

So... yeah, just some weird stuff I noticed. With this particular source, I noticed this problem with almost every de-interlacing method I tried and the same difference when altering first-field order...
Reply
Thanks given by:
#2
Any chance you are doing a conversion from interlaced YV12 to another colorspace somewhere before deinterlacing?
Reply
Thanks given by:
#3
That's a valid question, however I NEVER do anything with interlaced video before I de-interlace it. That's always my first step. The interesting thing here (at least to me) is that I have also tried to de-interlace it by way of doubling the frame rate, where every field ends up being it's own frame and I can see this same damage in some of the fields. Therefore, the only way to get rid of those damaged frames, is to de-interlace it twice (both times from source file) where the first is proper "top-field-first" and the second time "bottom-field-first"...

Personally, I suspect this has something to do with the fact that it's MPEG2 video at a lower bitrate, as I have seen this problem in other MPEG2 sourced videos as well.
Reply
Thanks given by:
#4
(2015-06-01, 11:23 PM)jerryshadoe Wrote: That's a valid question, however I NEVER do anything with interlaced video before I de-interlace it.

When I said "you" I really meant your workflow, not necessarily you deliberately adding a conversion. Many years ago, all MPEG-2 source filters for Avisynth would convert to YUY2; there was no native YV12 output and if you weren't careful this sort of messed up chroma was the result. Avidemux may still have this limitation.

Perhaps you can clip a little bit of the source where the problem is clear so we can test.
Reply
Thanks given by:
#5
Sorry I didn't respond sooner but I have been so busy that I haven't even been home the last couple of days...

To the best of my knowledge, using Avidemux 2.6, there is support for YUV2 and YV12 output and the setting HAS to be properly adjusted for output in the codec settings. I am using lagarith for this, and it has an option for both and RGB. I am assuming that the conversion happens properly, as when I look at a frame from the interlaced source, it has identical chroma/luma to the de-interlaced material.

I am happy to provide a test clip, if necessary, as soon as I catch up on everything... However, I figured out what the problem with my "something evil" footage. After ~56 minutes in, the source "switches" from top-field-first to bottom-field-first while still being encoded as top-field-first. The way I figured this out was the fact that a different encode, from a different station (and country) but using the same transfer (I checked the framing, timing, colors and all are the same; including film scratches) has the EXACT same problem that starts at the same time. Unfortunately, the other source was already de-interlaced (by someone who wasn't paying attention) and the footage is fine until it reaches that mark, where the field order shifts, resulting in some very nasty comb lines along with blended fields. Luckily, in the interlaced footage, I was able to catch this and work around itWink
Reply
Thanks given by:


Forum Jump:


Users browsing this thread: 1 Guest(s)