H.264 suited for SD DV home videos?

[jfcarbel]

All search provided on this topic was this post (http://forum.doom9.org/showthread.php?t=142007&highlight=h.264+interlace) which was rather inclined to de-interlace.

My question is based on this understanding, I was told at one time that MPEG-2 was better suited for encoding DV video then H.264 since MPEG-2 can encode interlaced content without de-interlacing or other motion estimation enhancements.

Can H.264 provide same PQ as MPEG-2 encoder for interlaced SD DV video content (http://en.wikipedia.org/wiki/DV) without adding motion artifacts to video.

So I wanted this thread as a discussion to clarify this point.

So here we go:

I know that MPEG-4 Main profile states support for interlaced material. But how exactly does that compare with MPEG-2 encoding of interlaced material and what exactly does "support" mean.

I have always been told that MPEG-2 is best for my DV SD resolution home videos because it handles compression best for this content and keeps the interlacing without issue. When I inquired a year ago if I could use MPEG-4 for my DV video with comparable quality results and where encoder would not try and de-interlacing, I was told it did not handle interlaced material as well as MPEG-2 encoder.

Is this true?

Currently I use ProCoder which is said to be the best for quality MPEG-2 encodes for DV material. And I do like the results. But wanted to see if I could get same quality from newer codec for same content at smaller file sizes.

Someone had mentioned in ProCoder forums:
Procoder absolutely shines with complex interlaced footage such as DV. It has a field based encoding function specifically for this

But on the doom9 forum I found this which seems to indicate MPEG-4 standard should support this. But whether the x264 encoder does is another question:
For interlaced content, the H.264 standard allows two fields to be coded either jointly, i.e. frame-based coding, or separately, i.e. field-based coding [1]. The frame/field coding concept can be extended to the macroblock level called Macroblock-Adaptive Frame/Field (MBAFF) coding in H.264. The concept of macroblock frame/field coding decision was originated from MPEG2 standard. Instead of splitting up a 16x16 MB into two 16x8 blocks, super MB [2] is defined as a decision unit. Each super MB consists of two vertically adjacent MB’s. The advantage of super MB is that all 7 block sizes can be used in either frame or field coding. Macroblock-Adaptive frame/field coding provides additional gain over picture-level adaptive coding

Other then x264 encoder, are their commercial ones that may support interlaced MPEG-4 encoding and offer similar quality to that I see in ProCoder?

If x264 does indeed support interlaced encoding without doing any destructive de-interlacing techniques first, then what are the options I need to use (i.e. width of the encoded file to be mod16, interlaced flag, MBAFF flag, etc)?
Am I basically just looking for an H.264 encoder that supports MBAFF? Or do I need one that supports PicAFF? Or is it more then that?

UPDATE:
Looks like x264 does support interlaced encoding using the options (tff/bff) and uses MBAFF. However, I have read that Proper MBAFF interlaced encoders very expensive. I believe by proper, they might be refering to fact that x264 uses the MBAFF method but without the adaptive part. According to Wikipedia: x264 does not support Adaptive MBAFF (Macroblock-Adaptive Frame/Field[12]) for interlaced videos, thus sacrificing a lot of coding efficiency in interlaced mode. This is the only major feature missing from x264 that is provided in all the other major implementations. It is being developed as part of Google Summer of Code 2010

This seems to explain MBAFF and its changes to interlacing in the encoding process:
MBAFF allows an AVC encoder to examine each block in a frame to look for similarities between interlaced fields. When there is no motion the fields will tend to be very similar, resulting in better quality if you encode the block as progressive video. For blocks where there is motion from one field to another the quality is more likely to suffer if encoded progressive, so these blocks can remain interlaced.

I assume since the "adaptive" part is missing from x264 that it just encodes everything as interlaced. Would that be a correct assumption? Seems so according to this (http://wiki.videolan.org/SoC_x264_2010#Adaptive_MBAFF_support).
So the benefit of a smaller file size might be nullified by this over MPEG-2 file size. I will do some tests and report back.

[kieranrk]

Just deinterlace it with tgmc and encode it progressively. The deinterlacing will be far better than what any television could do.

[aegisofrime]

Just deinterlace it with tgmc and encode it progressively. The deinterlacing will be far better than what any television could do.

Agreed. I think ever since TGMC came out the idea of deinterlacing being "destructive" has itself been destroyed ;)

[jfcarbel]

Just deinterlace it with tgmc and encode it progressively. The deinterlacing will be far better than what any television could do.

I guess I should clarify, I am looking for the best PQ. So whatever would give me that for my highly interlaced home videos, whether that be MPEG-2, de-interlaced H.264 encode, or interlaced H.264 encode. And if H.264 could offer very comparable quality to MPEG-2 yet save on file space, then that would be a benefit also. However, I am hearing that H.264 encoded with interlacing may be similar in file size to MPEG-2 anyways. I will need to do some tests and see.

On DV video (fully interlaced home video captures) I was told that H.264 cannot compete with MPEG-2 in PQ. Is this true is what I am trying to ask. I am no expert in this and frankly have many questions, thus my post here for the exports.

If I did go method of de-interlacing if that is what would provide comparable PQ in H.264 to that of MPEG-2 interlaced, then I would go with the slower but more quality: yadmifmod()+nnedi() or mcbob()+nnedi() de-interlace filters for AVIsynth. Processing time is not a factor since I am aiming for highest PQ.

[kieranrk]

On DV video (fully interlaced home video captures) I was told that H.264 cannot compete with MPEG-2 in PQ. Is this true is what I am trying to ask.


No this is not true.

The highest quality method would be tgmc in "bob" mode.

[jfcarbel]

No this is not true.

The highest quality method would be tgmc in "bob" mode.

Oh I see tgmc is TempGaussMC (http://avisynth.org/mediawiki/TempGaussMC) which does use yadmifmod() and nnedi()

Also my target display for these would be a HDTV and not a computer or old interlaced CRT. So maybe based on that target it is best to deinterlace now then having interlaced H.264 content then deinterlaced by the HDTV which I assume it would.

I wonder if once x264 gets the "adaptive" part done in MBAFF interlaced encoding, if it will surpass or equal the quality of TempGaussMC method. It looks like proper (adaptive) MBAFF will avoid trying to deinterlace the frames if not possible for good PQ and just encode as interlaced. But I would assume with TempGaussMC its an all in method meaning every interlaced frames get combined which I assume could be problematic in some cases (fast panning, quick motion). However, most of my home videos have little movement except for the shaky cam effect which I assume might give problems for deinterlacing as well. Maybe a de-shake filter exists.

[JEEB]

/me points at QTGMC (http://forum.doom9.org/showthread.php?t=156028) that actually supports NNEDI3 and friends, which is a speed-up in itself thanks to the new asm optimizations.

And no, most probably no TV will ever get on the level of TGMC or QTGMC in the close future. Also, you are mistaking deinterlacing for inverse telecine or just telecine if you think that deinterlacing should be "sticking the fields together". See this classic blag post (http://mod16.org/hurfdurf/?p=12) by TheFluff to know the two most basic types of content you can have on an interlaced stream (be it 60/1.001i or 50i).

Of course, some deinterlacers did actually do what you say, but the word "bobbing" was already mentioned here, which is (basically) to resize your fields to the height of a frame with as good of an algorithm as possible. Which is the best way to deinterlace currently.

[jfcarbel]

/me points at QTGMC (http://forum.doom9.org/showthread.php?t=156028) that actually supports NNEDI3 and friends, which is a speed-up in itself thanks to the new asm optimizations.

And no, most probably no TV will ever get on the level of TGMC or QTGMC in the close future. Also, you are mistaking deinterlacing for inverse telecine or just telecine if you think that deinterlacing should be "sticking the fields together". See this classic blag post (http://mod16.org/hurfdurf/?p=12) by TheFluff to know the two most basic types of content you can have on an interlaced stream (be it 60/1.001i or 50i).

Of course, some deinterlacers did actually do what you say, but the word "bobbing" was already mentioned here, which is (basically) to resize your fields to the height of a frame with as good of an algorithm as possible. Which is the best way to deinterlace currently.

Thanks for that link, I will give it a read.

So what does it mean when it says that MBAFF encodes the interlaced as progressive?

Wow, I am more confused now and my head is "bobbing" :)

... resize your fields to the height of a frame with as good of an algorithm as possible. Which is the best way to deinterlace currently

Is this what TGMC does? Or a combination of TGMC plus something else I would need to do?

I am open to using other encoders that might do proper MBAFF (commericial included) if it keeps it more simple and quality can compare to ProCoder MPEG-2 output on my SD DV NTSC home videos.

[kieranrk]

It looks like proper (adaptive) MBAFF will avoid trying to deinterlace the frames if not possible for good PQ and just encode as interlaced. But I would assume with TempGaussMC its an all in method meaning every interlaced frames get combined which I assume could be problematic in some cases (fast panning, quick motion).

MBAFF doesn't do any deinterlacing. It's a coding method that allows the encoder to encode a block as progressive (e.g for still scenes)

tgmc and variants are full rate deinterlacers which convert two fields into two frames which will give you full temporal motion.

[7ekno]

.. and quality can compare to ProCoder MPEG-2 output on my SD DV NTSC home videos.

The point that is trying to be made (that is eluding you) is that the encoder is not the "limiting" step for "quality" here ;)

"Quality" is bound by "whatever" is doing the "deinterlacing" ...

If your DVDs are played back on a 1080p HDMI upscaling DVD player, then it is doing the deinterlacing (on the fly) ... if a standard DVD player is feeding a HDTV panel via RCA, then the HDTV is doing the deinterlacing (on the fly) ...

The point all the posters above are making is that a PC with 2-4 cores of dedicated computing power running at 5 FPS can do a much better job at deinterlacing than a weak single core APU in a HDTV/DVD player that is forced to do the conversion on-the-fly (at ~ 30 FPS) ...

So while MPEG-2 might reproduce the interlacing better than H.264 (which I actually doubt anyway), the device responsible for deinterlacing will just trash the detail anyway (so why not burn your CPU power doing the best deinterlacing job that you can and encode to H.264, the most efficient format on the planet, while burning CPU cycles anyway?!?)

7ek

[nm]

I am open to using other encoders that might do proper MBAFF (commericial included) if it keeps it more simple and quality can compare to ProCoder MPEG-2 output on my SD DV NTSC home videos.

Even x264's current MBAFF implementation will give much better results (quality/bitrate) than any MPEG-2 encoder. I don't think adaptive MBAFF would help much further with shaky home videos that would be mostly encoded as interlaced anyway.

But as people have said, deinterlacing is what determines the final quality and it's much better to use TGMC than let the TV or some player use its simple motion-adaptive method. You can also filter the video further after deinterlacing. There's a decent deshaker for AviSynth, for example.

[SeeMoreDigital]

In my opinion if your source is pure interlaced you have two encoding options...

Option 1: You generate a pure interlaced encode at the same resolution as your source. Which is possible with MPEG-4 Part-2 ASP and MPEG-4 Part-10 AVC.

Note. Some direct-show decoders still don't properly support playback of pure interlaced encodes. Hardware players offer better playback support!


Option 2: You separate each pure interlaced field and encode it as a progressive frame (which doubles the fps speed).


Cheers

[jfcarbel]

In my opinion if your source is pure interlaced you have two encoding options...

Option 1: You generate a pure interlaced encode at the same resolution as your source. Which is possible with MPEG-4 Part-2 ASP and MPEG-4 Part-10 AVC.

Note. Some direct-show decoders still don't properly support playback of pure interlaced encodes. Hardware players offer better playback support!

Option 2: You separate each pure interlaced field and encode it as a progressive frame (which doubles the fps speed).

Cheers

Sounds like either way that H.264 can handle interlaced material as well if not better then high quality MPEG-2 encoders like ProCoder.

Option 2 sounds like it takes more processing time, but will yield better PQ.

For option 2, would I need to do any resizing of my DV 720x480 content and if so are their good guides on how to do this?
This is in reference to JEEB's comment:
Of course, some deinterlacers did actually do what you say, but the word "bobbing" was already mentioned here, which is (basically) to resize your fields to the height of a frame with as good of an algorithm as possible. Which is the best way to deinterlace currently.

Thanks for all the input. I will begin some encoding tests now using both ways and compare to ProCoder. Can I put some clips up here for judgement and the process I used when completed? Also any tips on using TGMC for my content would be greatly appreciated.

Also I generally use in ProCoder at MPEG-2 with avg bitrate="5500" max_bitrate="7500" min_bitrate="1000" in VBR encoding mode at mastering quality setting. What would be comparable bitrates in H.264 as I understand it can offer similar PQ at lower rates, but not sure if interlaced material effects this benefit.

ProCoder also applies some DNR to provide for less artifacts in the encodes, so what filter might be comparable to that? Maybe Smart Smoother HQ combined with MSharpen?
From what I've seen Canopus Procoder runs the video through a low pass filter. This reduces high frequency noise making the video a little easier to compress. It leads to slightly less pronounced macroblocks but also a less crisp image.

[jfcarbel]

Here is another question, some DV home video content I have is captured from older VHS tapes and the quality on these is often noisy and deteriorated even when using high end VCR with built in TBC. So for that content, would the deinterlacing process produce more artifacts due to the bad quality of the content and would it thus be better to just encode the content with interlacing?

I think the answer might be that it does not matter since it sounds like the TV will have to deinterlace it anyways and today's computer deintelacing algorithms are better.

For older tapes like this, I saw someone suggested this filter chain:

Video Denoise (default)
Dynamic Noise Reduction (7)
hue/saturation (1, .095, 1)
Smart Smoother HQ (3, 40, 160, 0)
brightness/contrast (+7%, 102%)
Flaxen VHS (default; horizotal +6)
Null Transform (cut off black and noisy edges)
MSharpen (80, 20)
Sharpen (8)

Where in this filter chain would I use TGMC? Always at the end maybe?

Also any opinions on Neat Video (http://www.neatvideo.com) for this type of cleanup versus using the filter chain above?

[nm]

Here is another question, some DV home video content I have is captured from older VHS tapes and the quality on these is often noisy and deteriorated even when using high end VCR with built in TBC. So for that content, would the deinterlacing process produce more artifacts due to the bad quality of the content and would it thus be better to just encode the content with interlacing?
It's better to deinterlace and filter before encoding. You can eliminate some of the artifacts and the result may have better visual quality. At least it will compress better.

http://avisynth.org/mediawiki/DeVCR
http://forum.doom9.org/showthread.php?t=144271

I think the answer might be that it does not matter since it sounds like the TV will have to deinterlace it anyways and today's computer deintelacing algorithms are better.

Yes. AviSynth also has much better denoising and sharpening filters.


For older tapes like this, I saw someone suggested this filter chain:
[...]
Where in this filter chain would I use TGMC? Always at the end maybe?

Always at the start. Most filters need progressive video to function properly. TGMC can also denoise the video while deinterlacing.


Thanks for all the input. I will begin some encoding tests now using both ways and compare to ProCoder. Can I put some clips up here for judgement and the process I used when completed?

Sure! That way you might get useful suggestions on how to improve the filtering chain. Include the source clips too, if possible.

Also I generally use in ProCoder a avg bitrate="5500" max_bitrate="7500" min_bitrate="1000" in VBR encoding mode at mastering quality setting. What would be comparable bitrates in H.264 as I understand it can offer similar PQ at lower rates, but not sure if interlaced material effects this benefit.

For archival purposes, I'd make very high-quality encodes without VBV limits. "Constant-quality" encoding at CRF=12 or less would be good, considering the HDD prices. Then you can make multiple lower-quality encodes for different devices by using either the high quality files or some lossless intermediate file to avoid filtering again:

1. source -> filtered lossless encode -> high-quality x264 encode to archive

2. lossless encode -> Blu-ray encode

3. lossless encode -> web encode

Bitrates depend on the content, which is also affected by filtering. Try CRF 18 with proper VBV limits and other settings for BD.

When comparing to the MPEG-2 encoder, make a few encodes with progressively smaller bitrates: same, 75 %, 50 %.

[nm]

Also any opinions on Neat Video (http://www.neatvideo.com) for this type of cleanup versus using the filter chain above?

You should probably open a thread about this on the Avisynth Usage subforum (http://forum.doom9.org/forumdisplay.php?f=33).
The filtering chain you quoted above seems outdated. More than 6 years old (http://forum.doom9.org/showthread.php?t=77262), perhaps? ;)

[jfcarbel]

You should probably open a thread about this on the Avisynth Usage subforum (http://forum.doom9.org/forumdisplay.php?f=33).
The filtering chain you quoted above seems outdated. More than 6 years old (http://forum.doom9.org/showthread.php?t=77262), perhaps? ;)

Good suggestion nm, I will do so. Thread was getting off topic a bit.

[Blue_MiSfit]

H.264 makes MPEG-2 look very old and outdated, which it really is. Interlaced coding doesn't change this equation either, and in the case of MBAFF, it makes H.264 look even better.

Don't get me wrong, MPEG-2 is still extremely useful, and isn't going ANYWHERE anytime soon. But don't let anyone troll you into thinking that H.264 is somehow less suitable for coding interlaced content. That's good for a laugh :D

Also, for archival purposes, there's nothing wrong with a low CRF, plus VBV to keep things under control. If you skip VBV, you'll probably have no hope of getting playback to work on a standalone / other hardware decoder. Even if you're not planning on using said devices, I've found it's better to err on the side of compatibility, especially since VBV shouldn't hurt quality much at all, provided you use something like HP@L3.1 for SD, which allows 17.5mbps, which should be more than enough bitrate for all but the most pathological sources. For HD, I'd suggest HP@L4.1, which allows up to 62.5mbps.

Derek

[jfcarbel]

H.264 makes MPEG-2 look very old and outdated, which it really is. Interlaced coding doesn't change this equation either, and in the case of MBAFF, it makes H.264 look even better.

Don't get me wrong, MPEG-2 is still extremely useful, and isn't going ANYWHERE anytime soon. But don't let anyone troll you into thinking that H.264 is somehow less suitable for coding interlaced content. That's good for a laugh :D

Also, for archival purposes, there's nothing wrong with a low CRF, plus VBV to keep things under control. If you skip VBV, you'll probably have no hope of getting playback to work on a standalone / other hardware decoder. Even if you're not planning on using said devices, I've found it's better to err on the side of compatibility, especially since VBV shouldn't hurt quality much at all, provided you use something like HP@L3.1 for SD, which allows 17.5mbps, which should be more than enough bitrate for all but the most pathological sources. For HD, I'd suggest HP@L4.1, which allows up to 62.5mbps.

Derek

I sorta thought H.264 could handle DV just as well as MPEG-2, thus my posting here now that I am getting started on capturing all my home videos and encoding them. I figured since H.264 can handle bigger bitates at lower file sizes then MPEG-2 that I could get more benefit and PQ from H.264 but at same file size.

According to this chart here (http://rob.opendot.cl/index.php/useful-stuff/h264-profiles-and-levels/), looks like my 720x480@30fsp NTSC SD video would be level 3. Did you suggest a level up just to get better archival PQ?

When level is selected do I even need to set a bitrate for the encoder? What CRF value/range would you suggest keeping in mind this is DV video 720x480?

Also back in this thread (http://forum.doom9.org/showthread.php?t=119719) you suggested encoding with interlaced might be better (of course that was a post from 2006). Do you still fell this way?
This thread also seems to indicate that the deshaker filter might add in the compression with less artifacts by removing shaky cam common to handholds for DV home videos.

[Blue_MiSfit]

I sorta thought H.264 could handle DV just as well as MPEG-2, thus my posting here now that I am getting started on capturing all my home videos and encoding them. I figured since H.264 can handle bigger bitates at lower file sizes then MPEG-2 that I could get more benefit and PQ from H.264 but at same file size.

You're getting terms confused. Bitrate and filesize are inherently linked. Bitrate * length = file size. This is true of any compression format, and MPEG-2 and H.264 have no differences in this regard. You're correct in thinking that you CAN get better picture quality with H.264 instead of MPEG-2 for a given source at a given bitrate. In fact, the quality difference will be enormous, unless you use equally enormous bitrates!

According to this chart here, looks like my 720x480@30fsp NTSC SD video would be level 3. Did you suggest a level up just to get better archival PQ?
Basically, yes :) If you code it interlaced, it will be 720x480i30. If you decide to deinterlace it, you will be encoding it at 720x480p60. For this, you will need Level 3.1.

When level is selected do I even need to set a bitrate for the encoder? What CRF value/range would you suggest keeping in mind this is DV video 720x480?
You need to set a CRF value OR a bitrate, regardless if you select a level or not. In fact, x264 will auto-set the level (as a best guess) to match your settings. So, if you feed it 480p60 video, it will auto select AT LEAST level 3.1.

I'd suggest starting at CRF 16, and if it's not totally transparent to your sources, step down to CRF 14 or so. Not much point in going lower than CRF 12 though. I'd suggest (assuming 480p60):


--level 31 --crf 15 --tune film --vbv-maxrate 17000 --vbv-bufsize 17000


Add in the rest like the source / output etc... also use the slowest preset you can tolerate.

Also back in this thread you suggested encoding with interlaced might be better (of course that was a post from 2006). Do you still fell this way?
This thread also seems to indicate that the deshaker filter might add in the compression with less artifacts by removing shaky cam common to handholds for DV home videos.

It really depends. You can try both :)

The big deal with deinterlacing before, and encoding at 60p is that you can use OMG_AWESOME deinterlacers like QTGMC. This adds to the encoding time, but overall picture quality will be better. Try encoding interlaced, and letting your GPU deinterlace on playback to 60p. If you're ok with its hardware deinterlacing, then just do that. If you prefer the QTGMC look enough to spend the time, then do that :)

It's all subjective, just like everything else in this damn world of video compression :devil:

Good luck,

Derek

[Bordo32]

I am getting intrigued, since I have the same task for my old home DV videos.
Even they are perfectly fine to archive as is - DV avi files on HDD, but the problem is that I cannot watch them directly by media file player such WDTV Live or etc.
If DV would be encoded to H.264 I may to get two targets: archiving and readiness for instant watching, and all with the retained original quality.

Jfcarbel, could you please update on your experiment results, interlaced vs. none interlaced H.264 encoding and etc.

[ronnylov]

...


--level 31 --crf 15 --tune film --vbv-maxrate 17000 --vbv-bufsize 17000


....

Derek

--level 31 is the same as --level 3.1 ?

And if level already is specified I thought that included the appropriate --vbv-maxrate and --vbv-bufsize or is it required to specify the VBV settings in addition to the level?

Regarding archival of DV i prefer to store the original DV avi files which already is compressed (they are smaller than lossless encoded files). But if I want to archive edited video or analogue captures I prefer x264 with crf around 13.

[audyovydeo]

Regarding archival of DV i prefer to store the original DV avi files which already is compressed (they are smaller than lossless encoded files). But if I want to archive edited video or analogue captures I prefer x264 with crf around 13.

Havent we all grappled with this problem !

I personally assume that if I feel something is important enough to be archived, it's because it's going to be edited later, somehow.
Edit = NOT lossy, so I keep it in DV. Certainly not H.264 lossless (why waste time converting ?)

cheers
a/v

[nm]

--level 31 is the same as --level 3.1 ?

Yep.

And if level already is specified I thought that included the appropriate --vbv-maxrate and --vbv-bufsize or is it required to specify the VBV settings in addition to the level?

You also need to specify the VBV settings. --level only sets the level flag.

[ronnylov]

Havent we all grappled with this problem !

I personally assume that if I feel something is important enough to be archived, it's because it's going to be edited later, somehow.
Edit = NOT lossy, so I keep it in DV. Certainly not H.264 lossless (why waste time converting ?)

cheers
a/v

If you by edit mean just cutting then I agree but if you also apply filtering, adjust levels, deshake and other stuff then it must be recompressed and if recompressing I prefer to recompress it to x264 instead of DV (which also is a lossy format but at a higher filesize for the same quality). Of course you may also store original DV files and the project settings of your edit as some sort of script (maybe an avisynth script) so you can reproduce the editing using the original files without loosing quality.

[jfcarbel]

Sorry I know this is an old thread, but lots of good info here and wanted to link it back to another "issue" with this project now that I am back at it.

My post at end of this thread (http://forum.doom9.org/showthread.php?t=166352)indicates the DV codec can also cause chroma issues and I am looking at suggestions to solve that.

So anyone that had done some DV to x264 as interlaced, speak up. TIA