Deinterlacing: double-framerate vs not doubleframerate?

[knutinh]

Some of the suggested parameters for deinterlacing puzzles me.

For any "real" interlaced content (not film-based telecined, but real video where every field is grabbed at separate moments of time), I cant see why anyone would want anything but 60p out of the deinterlacer when the input is 60i?

In theory, interlacing could give the sharpness of 30p, and the smooth movement of 60p (but not at the same time). Why would I drop even the possibility of those two benefits when deinterlacing?

If the problem is bitrate, surely there are other (more perceptually optimal) ways to reduce the bitrate out of the lossy video encoder?

[zilog jones]

Personally I keep "real" interlaced video interlaced, and either try and play it on an interlaced display if possible or bob deinterlace on playback (Yadif or whatever, might try out TGMC seeing Didee's recent posts about how fast it can go and still provide good results). I hate seeing 50i/60i material frame deinterlaced to 25p/30p, it looks awful IMO - I hate it especially when it has been done on a TV broadcast or DVD production, e.g. when showing a clip from some old TV programme (combined with 16:9 zoom for extra badness).

Sometimes it is a necessary evil, e.g. streaming video services like YouTube, Vimeo, etc. (most will not handle framerates higher than 30 fps, you could upload interlaced video but will have to extract the FLV/MP4/whatever and play it outside by other means); or encoding for some device that cannot handle neither interlaced video nor high framerates like most PMPs.

[knutinh]

But my point was that deinterlacing "true" 60i to 30p seems like misunderstanding the (potential) of interlaced video.

Native interlaced displays are rapidly vanishing from the picture, so deinterlacing is pretty much needed somewhere along the pipeline as long as the industry persist on using 50 year old analog compression technology

[wonkey_monkey]

Native interlaced displays are rapidly vanishing from the picture, so deinterlacing is pretty much needed somewhere along the pipeline as long as the industry persist on using 50 year old analog compression technology

Natively interlaced displays may be disappearing, but every TV knows what to do with interlaced video, and most do a pretty good job of displaying it. I've heard it said that interlaced video has, perceptually, 70% of the quality of progressive video at the same resolution (but with half the pixels). I'm not sure how that then affects the later stages of digital compression, but that still sounds like it makes a pretty attractive first step for broadcast compression.

But my point was that deinterlacing "true" 60i to 30p seems like misunderstanding the (potential) of interlaced video.

I think it's more "not caring" than "misunderstanding" - or people just don't want to deal with the problems anything about 30p causes (see above).

David

[wonkey_monkey]

That's for static areas. The idea is that ~30% of the resolution has been blurred away to prevent flickering. This also means that in motion, there is only ~35% the quality of progressive video.

I'm fairly sure that what I was reading was specifically talking about perceived overall quality, though what you say makes sense too. Is 1080i stuff still treated to prevent flickering, given that it'll be post-processed to progressive anyway?

Re: the ~35% figure, if the movement is 1 pixel per field, you'll lose the same line every time, but otherwise there's still extra detail in alternate fields. Assuming random motion, shouldn't we average that to (70+35)/2=52.5? And that only affects vertical motion, so do it again and you get 61.25%... (this is just me flinging numbers around, not chasing after a particular point).

I still think interlacing is a useful compression method. Those cases where it might perform poorly (motion) are exactly those cases where it's less likely to be perceived - same goes for other compression methods too, I suppose.

David

[2Bdecided]

In theory, interlacing is obsolete with an optimal encoder - if the encoder is that smart, throwing away half the information before it reach it is silly - heck, the encoder could even "interlace" internally if/when it wanted to.

In practice, it still has some use - encoders aren't "optimal" - and more importantly there's far more 1080i capable hardware (STB, BluRay etc) out there than 1080p50/60 - so if you want maximum temporal and spatial resolution at present, you must use 1080i.


Always double rate here - except for YouTube(!) - equally bemused why people do single rate, except for compatibility with certain hardware/software e.g. DivX on a stand-alone.

VLC bob (linear) is usually good enough for viewing - TGMC for processing etc.

Cheers,
David.

[knutinh]

The advantages and disadvantages of interlacing has already been discussed to great length in another thread:
http://forum.doom9.org/showthread.php?t=138426

One link in that thread caught my attention: When a bunch of people were asked to do side-by-side comparisions of 720p50, 1080i50 and 1080p50 that was compressed by h264 at a range of bitrates, they generally preferred 720p and 1080p over 1080i. What more do we need to know, in real-world scenarios, progressive looks better than interlaced if both are available. Interlaced should only be used as a legacy format for crappy equipment.

-k

[wonkey_monkey]

When a bunch of people were asked to do side-by-side comparisions of 720p50, 1080i50 and 1080p50 that was compressed by h264 at a range of bitrates

Out of curiousity, do you remember what was it displayed on?

Is there perhaps some advantage to 1080i in that it can be used to broadcast 1080p25 without repeating frames? Then a station can broadcast 50i and (effectively) 25p without switching formats...

David

[knutinh]

Out of curiousity, do you remember what was it displayed on?

Is there perhaps some advantage to 1080i in that it can be used to broadcast 1080p25 without repeating frames? Then a station can broadcast 50i and (effectively) 25p without switching formats...

David
Some Pioneer high-end 50" 1080p plasma (Kuro?). The test used the deinterlacer of that display (a weakness of the test in my view, but lets assume that Pioneer is a reasonable best-case for actual televisions out there).

-k

[Alex_ander]

Out of curiousity, do you remember what was it displayed on?

I'm curious too, did they use (the proper) interlaced scanning mode for 1080i? Otherwise it's just a test for destructive or less destructive real-time deinterlace types used in particular TV models. I don't think they developed 1080i keeping in mind progressive scan mode in receiver. Each format is best displayed as it was shot, absolutely different types of motion presentation to man's eye. 'A bunch' of experts did testing while developing both formats. An evident advantage of using 720p is ability to capture better stills in motion.

Is there perhaps some advantage to 1080i in that it can be used to broadcast 1080p25 without repeating frames? Then a station can broadcast 50i and (effectively) 25p without switching formats...

Unlikely it will look better than 1080i on displays with interlaced mode thrown away. They will recognize the format as 1080i and deinterlace it (some of them - with flickering, if it's in full sharpness and deinterlacer bobs by single field or like this).
As for 1080p (50/60) at production stage (cameras/studios/storage), it is already useful since both 1080i and 720p can be obtained from it without any artifact. It can't simply replace 1080i for wireless broadcasting since it always costs bandwidth (or bitrate/quality).

[knutinh]

It can't simply replace 1080i for wireless broadcasting since it always costs bandwidth (or bitrate/quality).
The test I referred to indicated that 1080p or 720p beat 1080i for the same bandwidth/bitrate. There are always some limitations in such tests, but I think that this one was realistic enough to use as an argument.

Now, for communication between sensor and storage in-camera, using high-cost h264 compression may not be cost-effective. For that single case, interlacing may have some merit if one cannot afford the bandwidth and heat caused by doubling the pixel rate.

For most of the cases where we consumers are conserned, one can afford high-quality digital video compression, and inserting stone-age analog compression in series with that digital compression seems to cause cause nothing but head-aches, reduced quality and increased cost.

But this discussion is going off-topic. There is a thread for pros and cons of deinterlacing. I suggest staying on topic as that makes this forum a lot more informative for the reader:
Debate: Interlaced vs. Progressive (http://forum.doom9.org/showthread.php?p=1358769#post1358769)

In that thread, 2Bdecided posted the test that I have been referring to:
A Novel Method for Subjective Picture Quality Assessment and Further Studies of HDTV Formats, IEEE TRANSACTIONS ON BROADCASTING, VOL. 54, NO. 1, MARCH 2008, Hans Hoffmann, Member, IEEE, Takebumi Itagaki, Member, IEEE, David Wood, Tobias Hinz, and Thomas Wiegand, Member, IEEE (http://ip.hhi.de/imagecom_G1/assets/pdfs/ieee_sub_quality_hdtv_2008.pdf)

32 "experts" and 135 "non-experts" were included in the test.

The practical tests of the method have reinforced the conclusions of earlier investigations [9] on the use of 1080p/50, 720p/50 and 1080i/25, which found that a progressive HDTV format provides better failure characteristics on image quality than the 1080i/25 format when compressed with H.264/AVC and viewed on FPDs at typical broadcast bit rates between 6 and 18 Mbit/s.
...
With decreasing bit rates the resolution advantage of 1080p/50 was masked by compression artifacts, and the assessors voted in favor of
720p/50. Overall, and with the configurations used in this test, 720p/50 was clearly the most favorable format among the threeformats under test.
...
On displays: to date no large Grade 1 reference matrix display with precise electro-optical transfer characteristics and known deinterlacing performance is available. This may have contributed to the poorer results for the 1080i/25 format. However, the authors believe that the main reason is to be found
in the difficulties with the H.264/AVC encoding of 1080i/25 (because it contains only half the vertical-temporal information compared to the progressive formats), and with the interlaced ’footprint’ from the content source


-k

[wonkey_monkey]

Unlikely it will look better than 1080i on displays with interlaced mode thrown away. They will recognize the format as 1080i and deinterlace it (some of them - with flickering, if it's in full sharpness and deinterlacer bobs by single field or like this).

Surely any TV worth it's salt will analyse the picture to determine if it's truly progressive and process it accordingly?

David

[Alex_ander]

Surely any TV worth it's salt will analyse the picture to determine if it's truly progressive and process it accordingly?

David

I think it must recognize 24p/23.976p by framerate (as defined in BD standards), but not 25p. It is technically easy to flag such content, but do the standards demand it from broadcasters or TV manufacturers? Analysing video on-the-fly doesn't look as easy and reliable.

[Alex_ander]

@knutinh
Thanks for the link, just looked through that paper. As expected, they compared deinterlaced (=distorted by definition) and originally progressive, with predictable result (without interlaced scanning in receiver 720p must win). However, specific compression problems with interlaced also do matter (at least with h264 with its long GOPs and at low bitrates).
As for comparison 1080p(50/60) vs 720p, it's a question of simple proportion bitrate/quality. If you want it in higher resolution, you have to pay with broadcasting bandwidth. If you simply increase the resolution and leave bitrate value optimized for 720p, it may look better from the second row, but may become unacceptable for archiving or editing.

[zilog jones]

Surely any TV worth it's salt will analyse the picture to determine if it's truly progressive and process it accordingly?

Mine does attempt this (It's a Toshiba 32", forget the model but it's a mid-range set from about 2007), and I'm sure most others do with 25i/25p. If there's little motion between fields it will not deinterlace, but it doesn't always get it right - often with news reports when there's just some talking head it will not deinterlace resulting in combing (especially noticeable if something small moves in the background), and if there is small interlaced video overlayed on 25p footage it may not deinterlace. I've seen much worse results with other TVs though, where e.g. it won't deinterlace tickers on news channels resulting in an illegible mess.

...except for compatibility with certain hardware/software e.g. DivX on a stand-alone.
576i and 480i Xvid works fine on my old Yamada DVD-6500X (Zoran Vaddis 776 chipset), though I forget how it handles field order (it might only accept BFF or something). I guess with newer HD upscaling players it may be less likely they can output interlaced SD resolutions properly.

Also, Yadif was added to VLC recently (single and double rate).

[knutinh]

Surely any TV worth it's salt will analyse the picture to determine if it's truly progressive and process it accordingly?

David
Evidently, converting "interlaced" to progressive is a pattern-matching problem where:
1. No flags or metadata can generally be trusted
2. One has to classify the content as either "film" (telecine from a progressive source) or "video" (true interlaced), then apply appropriate processing for either.
3. Nvidia and ATI has hundreds of streaming processors and probably thousands of software developers, yet they struggle to get this right and routinely "break" this functionality in some driver revisions (video playback is not their core business, though)

In addition, many content producers mix video content with telecined content, or do edits in video content that breaks the cadence, meaning that long-term pattern recognition fails.

My guess is that high-end deinterlacers use motion-compensation to align several fields, then do adaptive processing to find "best guesses" on lost information. This is a process that can add delay, certainly adds cost, and that cheapo no-name tv manufacturers almost certainly dont care about.

-k

http://en.wikipedia.org/wiki/Telecine#Reverse_telecine_.28a.k.a._inverse_telecine_.28IVTC.29.2C_reverse_pulldown.29
http://en.wikipedia.org/wiki/Deinterlacing#Motion_Compensation

[knutinh]

@knutinh
Thanks for the link, just looked through that paper. As expected, they compared deinterlaced (=distorted by definition) and originally progressive, with predictable result (without interlaced scanning in receiver 720p must win). However, specific compression problems with interlaced also do matter (at least with h264 with its long GOPs and at low bitrates).

Well, for practical purposes I would say that the ONLY side-by-side comparision of interlaced and non-interlaced material that interests me is displayed on a progressive display. I will never own an interlaced display again. Why would I care how the video looked on an archaic display device, predictably you can increase quality somewhat on a monochrome display if you drop the color channels and spends a little more bandwidth on the luma/luminance channel, but why bother? :-)

And even though you (and I) think that the result is predictable, some do not think so. Some think that interlacing->deinterlacing are good pre/post-processing tools to increase the picture quality/bitrate of lossy compressed video. I have seen a Philips paper that tried to support this for the case of MPEG2.

The fact that we dont know what algorithm Pioneer use for deinterlacing is a weakness of this test. Concievably, newer, more advanced deinterlacing algos could give better results for the 50i alternative.

As for comparison 1080p(50/60) vs 720p, it's a question of simple proportion bitrate/quality. If you want it in higher resolution, you have to pay with broadcasting bandwidth.
Exactly. So we need to leave the PR-dogmas and concentrate of perceived quality instead.

-k

[2Bdecided]

Don't people believe that, given near unlimited bitrate and near unlimited processing, 1080i gives better quality that 720p?

The question is what happens in the real world.

Maybe broadcasters choose 1080i due to pure bloody mindedness - or because it's a bigger number - or perhaps they just want "sharp" HD - even if it does fall apart when things move?

Or maybe with typical encoders, content, and displays, it looks as good as, if not better than, 720p?

Cheers,
David.

[wonkey_monkey]

2. One has to classify the content as either "film" (telecine from a progressive source) or "video" (true interlaced), then apply appropriate processing for either.

Or non-telecined progressive source, such as we get here in the UK (films sped up to 25p, or just native stuff). As I understand it TVs (the good ones, anyway) do have something called cadence detection, which is for finding these patterns to determine what to do with the video.

My TV (a Samsung, which have always impressed me with their upscaling of video content) does sometimes get confused if there's a "film" rate background with a video foreground, but only if I've enabled the blur reduction function (which only seems to work with news tickers, credits, etc) - for example on QI, or the local news. The result isn't combing, but something else I can't quite describe - almost like reversed field order, but not quite.

David

[knutinh]

Don't people believe that, given near unlimited bitrate and near unlimited processing, 1080i gives better quality that 720p?

I believe that given:
-lossless compression
-a scene that coregisters every object in at least 2 consequtive fields with only translatory motion in-between where the y-component of the motion vector is 1,3,5,... lines
-a perfect deinterlacer that can track/compensate that movement
-an interlacer that does no vertical filtering before decimating

That 1080i50 can be identical to 1080p50. Problem is, that is a lot of givens. Many of which are totally unrealistic.

The quality of 1080i50 vs 720p50 given more realistic case is not easy to predict. I believe that the test that I quoted, shows uncompressed 720p50 to get a marginally better rating than 1080i50 (although significance levels overlap). That is with the Pioneer Kuro 50" plasma (using built-in deinterlacer) at a distance of 3 display heights and averaged over a set of test-sequences and 135 non-experts as well as 32 experts.

The question is what happens in the real world.

Maybe broadcasters choose 1080i due to pure bloody mindedness - or because it's a bigger number - or perhaps they just want "sharp" HD - even if it does fall apart when things move?

Or maybe with typical encoders, content, and displays, it looks as good as, if not better than, 720p?

Cheers,
David.
Perhaps they want to do lossless editing and archiving, and the interfaces available at the time did not allow 1080p50, while 720p50 was voted down because it was a worse PR-number, and because telecined movies look better in 1080i50 than 720p50?

-k