Debate: Interlaced vs. Progressive

[Vchat20]

So this has been one mind-bender that has been nagging at me for a while and that has been the battle between Interlaced and Progressive content and the supposed 'need' to deinterlace said interlaced content from the getgo in many encoding situations.

Here's my take on it: If the original content is already progressive, there really isn't much to do there. And there really is no way to 'interlace' it back again since theorhetically you'd just end up with identical frames anyhow.

If the original content was interlaced on the other hand, you are at the maximum of quality in terms of retaining as much information as you can. Deinterlacing inherently loses information even with the best deinterlacing filters. In addition, most playback devices/software at least with good built in filters are capable of doing good deinterlacing on the fly during playback. Many software dvd codecs are a good example of this.

With this all said and done, why do you suppose it has long been such a habit of deinterlacing every piece of content regardless of destination? What are your opinions on the matter.

FYI: This is not saying 'OMFG you must stop deinterlacing', but more of a start of a debate as to why you think this is still commonplace.

[GodofaGap]

I think deinterlacing became popular for three reasons:

- good deinterlacers were too hungry on the CPU to be be used in real-time back then and some new ones are now as well
- some codecs didn't have an interlaced mode (DivX 3.11 for example)
- a lot of times videos were resized to arbitrary sub-SD resolutions.

One method of deinterlacing doesn't throw away any information at all BTW: bobbing.

I agree with you that if your target is DVD on TV then deinterlacing doesn't make much sense. Also remember that a lot of 'deinterlacing' going on now in the AviSynth forum is actually trying to undo bad standard conversions.

[Mug Funky]

a lot of editors and directors want a "filmic" look (i don't like that word, but there aren't any others to use), so they deinterlace.

my personal opinion is if you want it to look like film, shoot it to look like film, or just cough up the dough and actually shoot film. a "film look" is not at all a post decision - it must be considered from before shooting starts. the video look comes from more than just the interlace.

as far as deinterlacing everything in the encoding world, it's probably because the majority of codecs don't support it, and a large amount of the delivery methods out there don't support it (you can't bob in youtube or flash).

i think video can look quite good and don't see the point in making something juddery and blurry/jumpy in the name of some impossible (in most video situations) film look.

[Revgen]

When it comes to sports footage, deinterlaced (bob-deinterlacing anyway) is always better IMO. Even if information is lost or none gained at all, the smooth motion that results from deinterlacing is pleasing on the eyes. And since sports footage always has a lot of motion, real-time bobbers are often not up to par and the sharp diagonal lines that make up the court or field can be aliased quite easily using conventional deinterlacers. Using a slow motion-based bobber like MCBob combined with NNEDI often is the best way to make the sports footage look good when it's deinterlaced.

[J_Darnley]

You must deinterlace when watching on a progressive display or you see the combing. When you do it is up to you. You can do it in the decoder, in pre-processing, in the second decoder, in post-processing. I do it because I watch everything on my PC using a progressive CRT monitor on it. If I made a DVD or something to be shown on my TV then I may leave it interlaced.

[Blue_MiSfit]

Right, but you're still at the mercy of your DVD player / TV's processor. Who knows, it might do a dumb bob. Obviously if you're encoding a DVD it's best to leave it interlaced, but if you're encoding a file for playback on a PC connected to a progressive display, I find it necessary to encode 60p. This is one reason I really like 720p!

[knutinh]

Interesting (old) thread. Basically we have a set of limitations. Camera sensor. codec. Display device. pre/post processing.

You will never get a higher field-rate or spatial resolution than that offered by the sensor. You will never get to watch a higher frame/field-rate or resolution than that offered by your display device. And in a bandwidth-limited world, your quality may be limited by the lossy compression, codecs may be more efficient on p than i material.

I tend to see interlacing as an analog 2:1 perceptually motivated compression method. I dont really see its purpose in this digital era, except for legacy purposes. If you want to trade motion/resolution/bandwidth, then use a lossy digital codec that does it intelligently..

-k

[Dr.Khron]

I've recently been watching a lot of football at my buddy's house, it looks GOOD on a modern plasma TV.

Am I crazy, or does the live (US NFL) football coverage constantly switch back and forth bewteen interlaced and progressive?

It often seems like you are seeing progessive graphics overlayed on an interlaced video stream, and some video streams seem smoother then others.

[Blue_MiSfit]

Yep, it's a real nightmare with live video - espeically when progressive graphics are overlaid on top of an interlaced video stream, with occasional cuts to a (probably) progressive slow-mo camera, and everything else, plus probably some upscaled SD - then toss in commercials!! :devil:

ROFL!

Once they start broadcasting 1080p60 this will all go away hehe

~MiSfit

[Dark Shikari]

I've recently been watching a lot of football at my buddy's house, it looks GOOD on a modern plasma TV.Some channels are 720p60, like ESPN, for example.

[Blue_MiSfit]

Exactly, and how nice is that? Especially considering their craptacular encoders ;)

Is anyone actually broadcasting H.264 in the U.S.? Cable / Satellite/ OTA?

~MiSfit

[Guest]

Is anyone actually broadcasting H.264 in the U.S.? Cable / Satellite/ OTA? Ever heard of DirecTV?

[kieranrk]

Exactly, and how nice is that? Especially considering their craptacular encoders ;)

Is anyone actually broadcasting H.264 in the U.S.? Cable / Satellite/ OTA?

~MiSfit

NBC and ABC uplink feeds and probably a few others. Dish as well.

[Ghitulescu]

While most sources are interlaced and most viewing devices are progressive - it's hard to answer.
NB: some progressive camcorders store their movies interlaced, for compatibility reasons. Probably the best example is Canon.

[Manao]

I tend to see interlacing as an analog 2:1 perceptually motivated compression method. I dont really see its purpose in this digital era, except for legacy purposes.And even so, what legacy ? Is there an analog 1080i ? No. So there is no legacy to preserve here.

If you want to trade motion/resolution/bandwidth, then use a lossy digital codec that does it intelligently..Especially since the tradeoff is actually worse than people think. On the paper, interlaced may sound good : you get the full vertical resolution when there are no motion, and the full temporal resolution when it moves. So 1080p60 and 1080i60 are supposed to be comparable, with 1080i60 saving perhaps 25% bitrate after compression, and reducing the decoding needs.

That's on paper only. As it happens :
you don't get the full vertical resolution. Oh, sure, there are 1080 row of pixels, so when the video is still, you're supposed to look at a 1080p video. And you do. Except that video has been downpassed vertically, so you are actually looking at a content that only has 600 or so rows of pixels of actual information.
You only decode 30 frames per seconds instead of 60. Yeah, sure, but you get the privilege to decode mbaff. On the paper, it's 'just' a coding tool that doesn't increase computational decoding complexity. But it does. Strongly. So indeed, it won't reach the computational needs of 1080p60, but there isn't a 2:1 margin either.
You're sending an interlaced signal to the TV, so somebody has to deinterlace it. Guess what, deinterlacing isn't cheap. When accumulated with the cost for mbaff, I think we reach the computational cost of 1080p60. But I cheat a bit here, because for legacy purpose, you would have needed a deinterlacer for SD content (but not for HD)


Now, I may be biased on the subject, and I might miss some arguments in favor to interlacing. But I don't see which ones.

[roozhou]

While most sources are interlaced and most viewing devices are progressive - it's hard to answer.
NB: some progressive camcorders store their movies interlaced, for compatibility reasons. Probably the best example is Canon.

And Sony.
It seems Sony's LCD TVs have quite good deinterlacer. I guess Sony will stick to producing interlaced Camcorders because it looks better on their LCD TV than others.

[tony uk]

the reason picture is better here is because of the interlaced
content poor people in usa only have NTSC : never the same colour.
merry christmas from uk:stupid:

[Blue_MiSfit]

Dish Network can't be broadcasting entirely H.264! My parents still have a receiver from at least 6 years ago that works perfectly. It's definitely MPEG-2 only!

I know they plan to broadcast H.264 in the future, or maybe they're only doing it for HD feeds? I wonder how these companies address legacy receivers.

~MiSfit

[Guest]

Dish Network can't be broadcasting entirely H.264! My parents still have a receiver from at least 6 years ago that works perfectly. It's definitely MPEG-2 only!

I know they plan to broadcast H.264 in the future, or maybe they're only doing it for HD feeds? I wonder how these companies address legacy receivers.
I don't know about Dish, but in the case of DirecTV, they initially were MPEG2 for everything (after an early MPEG1 phase!) including HD. Then they phased in the AVC HD channels. They required subscribers wanting HD to upgrade their receivers. For a short interim period you could still receive some HD channels in MPEG2 but they were eventually all converted to AVC.

I have an old MPEG2 only HD DirecTV box that I had to eat.

[Blue_MiSfit]

Interesting! I would love to see a proper comparison between Dish, and DirecTV's video quality - maybe with Comcast thrown in for good measure (as I know they still use a lot of MPEG-2 in most areas).

Do you know of such a resource, neuron2?

Also, how would you subjectively grade DirecTV's HD services?

Man - we're getting OT though! Looks like that's my fault :( .

To get things back on track - has anybody actually seen a 1080p60 broadcast in any form?

~MiSfit

[knutinh]

And even so, what legacy ? Is there an analog 1080i ? No. So there is no legacy to preserve here.

Broadcast, storage and interfaces usually does not support 1080p50/60. In other words, 720p60, 1080p30 and 1080i60 are the available options.

Especially for 24fps movie content, the case could be made for a 60i container. But it introduce an infinite list of possible screw-ups for engineers and content producers...

Especially since the tradeoff is actually worse than people think. On the paper, interlaced may sound good : you get the full vertical resolution when there are no motion, and the full temporal resolution when it moves. So 1080p60 and 1080i60 are supposed to be comparable, with 1080i60 saving perhaps 25% bitrate after compression, and reducing the decoding needs.

That's on paper only. As it happens :
you don't get the full vertical resolution. Oh, sure, there are 1080 row of pixels, so when the video is still, you're supposed to look at a 1080p video. And you do. Except that video has been downpassed vertically, so you are actually looking at a content that only has 600 or so rows of pixels of actual information.

I have heard information to the contrary: SD interlacing includes a vertical lowpass filter, while HD interlacing should not. I do not claim to know this for a fact.

I would have guessed that a high-end interlacer could be content-adaptive, filtering only moving parts of the scene?

Anyways, using interlacing as an extra layer of lossy compression makes little sense. If interlacing is/was a good way of removing bits while keeping quality, then MPEG/ITU codecs would do interlacing internally on progressive signals. And then there would be complete end-to-end control of what had been done and how it should be converted. The same can be said about colorspace conversion and decimation, though.

You're sending an interlaced signal to the TV, so somebody has to deinterlace it. Guess what, deinterlacing isn't cheap. When accumulated with the cost for mbaff, I think we reach the computational cost of 1080p60. But I cheat a bit here, because for legacy purpose, you would have needed a deinterlacer for SD content (but not for HD)

Sony and Philips have invested heavily in deinterlacing. One might suspect that they have an interest in keeping legacy formats that other companies does not do equally well.

I remember reading a Philips paper in which they compared 25p, 50i and 50p when encoding as bitrate-constrained MPEG2. The conclusion was that MPEG2 + 50i + HQ deinterlacing had best rate-distortion characteristics. Perhaps because MPEG2 lacked deblocking?


Now, I may be biased on the subject, and I might miss some arguments in favor to interlacing. But I don't see which ones.
There is a case for interlacing in sensors. If you are bandwidth/heat-constrained, then 60i may be better than 30p, especially if you can tailor OLPF and deinterlacing for the task. I would deinterlace as early as possible though.

There are some physical stuff in sensors that I do not understand very well. Integration time and bandwidth for instance.

[Manao]

I have heard information to the contrary: SD interlacing includes a vertical lowpass filter, while HD interlacing should not. I do not claim to know this for a fact.That's interesting. I must admit that I noticed there was a lowpassing when I took a 720p50 content and tried to turn it into a 576i50 one. It was unwatchable without (strong) lowpassing, thus I concluded that lowpassing was necessary for interlacing, and extrapolated the same was happening in HD. I was confirmed in that opinion when I read that document (http://www.ebu.ch/CMSimages/en/tec_svt_multiformat_v10_tcm6-43174.pdf), that explains how a 2160p video was transformed into a 1080i one (a 540p field is created by averaging four lines), but perhaps that document is incorrect.
I remember reading a Philips paper in which they compared 25p, 50i and 50p when encoding as bitrate-constrained MPEG2. The conclusion was that MPEG2 + 50i + HQ deinterlacing had best rate-distortion characteristics. Perhaps because MPEG2 lacked deblocking?There have been a lot of studies regarding 1080ixx vs 1080pxx. AFAIK, all of them but one (EBU's) concluded 1080i was better. I don't really know what to make of that, and I would have liked to see 720pxx thrown in the lot too.

[knutinh]

That's interesting. I must admit that I noticed there was a lowpassing when I took a 720p50 content and tried to turn it into a 576i50 one. It was unwatchable without (strong) lowpassing, thus I concluded that lowpassing was necessary for interlacing, and extrapolated the same was happening in HD. I was confirmed in that opinion when I read that document (http://www.ebu.ch/CMSimages/en/tec_svt_multiformat_v10_tcm6-43174.pdf), that explains how a 2160p video was transformed into a 1080i one (a 540p field is created by averaging four lines), but perhaps that document is incorrect.

Thanks for the link! Repeating it here for the discussion:

a) Interlacing to 1080i
For interlacing, every second 2164p-frame was shifted vertically two lines downwards. After deleting the first two and last two lines in the frames that were shifted (and the last four lines for the frames that were not shifted) to get 2160 lines again, each frame was filtered to 540 lines by line averaging (using Shake’s Box Filter). Horizontal filtering from 3840 to 1920 columns was performed using Shake’s Sinc Filter to benefit in perceived sharpness from the “oversampled” master. The two 540-line fields where then weaved into one single 1080-line interlaced frame. This process resembles the process in any video camera performing interlace in the basic default ‘Field Integration Mode’ – i.e. like a 2160 line video camera sensor reading out the
average of the sensor’s line 1+2+3+4 to Field 1; line 3+4+5+6 to Field 2; line 5+6+7+8 to Field 1 etc.

http://sci.tech-archive.net/Archive/sci.image.processing/2007-03/msg00026.html

Normally, if you have interlaced scanning with a video (NTSC) signal,
and do field integration, you put lines 1+2, 3+4, 5+6... together for
the first half-videoframe, and 2+3, 4+5, 6+7... for the second half-
videoframe.

This leads to a somewhat lower vertical resolution because of the
interpolation, but more than 240 lines.

In frame integration you normally get lines 1,3,5... for the first
half, and 2,4,6 for the second. So you get a better vertical resolution
(no interpolation), but you loose half of the charges what results in
higher noise.

Jens

http://www.damtp.cam.ac.uk/lab/digimage/cameras.htm

To complicate matters further, different cameras construct the two video fields in different manners. In some cameras the even field corresponds to the even lines of pixels in the CCD chip, and the odd field to the odd lines of pixels in the CCD chip.
...
Slightly better are cameras which produce an average of the even lines and the preceding odd lines for the even field, and the odd lines and the preceding even lines for the odd field.

To me it seems that both ways of producing interlaced content is feasible, and might(?) be found in different video cameras?

There will typically be an optical lowpass filter in front of the sensor that smears out details to some degree (have to show some respect to Nyquist) and optics seldomly have perfekt spatial frequency response either.

There have been a lot of studies regarding 1080ixx vs 1080pxx. AFAIK, all of them but one (EBU's) concluded 1080i was better. I don't really know what to make of that, and I would have liked to see 720pxx thrown in the lot too.
I think that finding a suitable "success metric" is going to be difficult and political.

Do you use PSNR, SSIM, or real people?

Do you average across all codecs and deinterlacer implementations (optimizing mean viewer experience), or only for some idealized reference implementation?

I think that 1080p60 with good lossy compression will be best on a quality vs bandwidth benchmark. But is that all? How do you factor in quality vs price?

BBC concluded that 720p was enough for the UK public as long as screen sizes did not go much beyond 50".

[MfA]

I know what to conclude, European standards agencies are superior ... it's like ITU vs MPEG, hardly a contest :p

[kieranrk]

BBC concluded that 720p was enough for the UK public as long as screen sizes did not go much beyond 50".

It's a shame that they once published information that actually made sense.


I have heard information to the contrary: SD interlacing includes a vertical lowpass filter, while HD interlacing should not. I do not claim to know this for a fact.


I always thought SD/HD were treated the same like Manao.

[MfA]

HD displays are all progressive, so the old type of flicker for static thin lines (fine text for instance) doesn't exist anymore when the deinterlacer works well (of course deinterlacers don't always work well). Motion dependent aliasing is actually more likely though for HD (whenever something moves near a multiple of 1 pel per field vertically the vertical bandwidth is halved).

A sports broadcast with interlacing and without a flicker filter would be interesting ... the grass would look lovely I bet.

[knutinh]

It's a shame that they once published information that actually made sense.

Do you think that the report does not make sense, or do you think that they have degraded since?

http://downloads.bbc.co.uk/rd/pubs/whp/whp-pdf-files/WHP092.pdf

-k

[2Bdecided]

All UK HD broadcasts are 1080i - there's no 720p - not "even" from the BBC.

And the official line is that while some old tests showed 720p looked better at lower bitrates than 1080i, encoders have improved and 1080i now looks better.

Cheers,
David.

[knutinh]

All UK HD broadcasts are 1080i - there's no 720p - not "even" from the BBC.

And the official line is that while some old tests showed 720p looked better at lower bitrates than 1080i, encoders have improved and 1080i now looks better.

Cheers,
David.
Do you have sources on this?

For 24p-originating content I think that it is feasible to prove that 1080i is "best" for most display devices. I think that it is difficult to prove the same for general content (e.g. sport) given that there are many bad deinterlacers out there. My VideoSeven lcd contains one of them :-)

The question includes "interlacers", lossy encoders, bitrate, and deinterlacers/scalers. Many variables...

-k

[scharfis_brain]

but most broadcasters just take a 1080i feed and bob-deinterlace it to 720p.
Their deinterlacers aren't the best either (staristepping etc.).
Then you display will upscale this again to 1080p.

I think the content should be left in its original format.

[2Bdecided]

Anyways, using interlacing as an extra layer of lossy compression makes little sense. If interlacing is/was a good way of removing bits while keeping quality, then MPEG/ITU codecs would do interlacing internally on progressive signals.Of course they don't, even though interlacing does (at least partly) achieve the gains it's supposed to. That's why it's used. It's not a conspiracy, and it's not a mistake - it actually works (i.e. gives better quality / lower bitrates). Even with H.264 (if the encoder handles interlacing well enough).

The other reason is that 1080 is a bigger number than 720 (and does look sharper on most TVs - even the previously common 768-line ones) - but the technology isn't out there to do 1080p50 yet, so you're stuck with interlacing.


It does make logical sense that packaging the (adaptive) interlacing and (adaptive) deinterlacing into the encoder should make it work better than externally - but it's more complexity: more tuning in the encoder; more work in the decoder. Has anyone ever done it?

Cheers,
David.

[Manao]

It does make logical sense that packaging the (adaptive) interlacing and (adaptive) deinterlacing into the encoder should make it work better than externally - but it's more complexity: more tuning in the encoder; more work in the decoder. Has anyone ever done it?It would be easier. It moves : resize and encode in 720p. It's static, keep 1080p and halve the framerate. You could even do that somewhat adaptively per macroblock. It would still be easier than mbaff.

[knutinh]

Of course they don't, even though interlacing does (at least partly) achieve the gains it's supposed to. That's why it's used. It's not a conspiracy, and it's not a mistake - it actually works (i.e. gives better quality / lower bitrates). Even with H.264 (if the encoder handles interlacing well enough).

I am not suggesting that it is a conspiracy, I am using it as an argument that you are wrong :-) Can you offer some references that h264 with interlacing has better PSNR/SSIM/subjective quality than h264 without?

For your statements to be generally right, I think one would expect that compressing any original 1080p50 sequence at:
1)1080@50p, h264, X mbps
2)1080@50i, h264, X mbps
3)720@50p, h264, X mbps

Would (on average) be best for 2) for any bitrate X. I highly doubt that to be true, but I have read Philips white-papers suggesting that they could make make 2) true if they used:
A)Philips' advanced deinterlacing
B)MPEG2 without deblocking filtering
C)At constrained bitrates

I think that B) was suggested as an important explanation.

The standardization organs are competitive about compression gain. If integrating interlacing/deinterlacing in the codec resulted in improved PQ for a given bitrate and a given implementation cost, surely someone would suggest it, have it implemented in the standard?

It does make logical sense that packaging the (adaptive) interlacing and (adaptive) deinterlacing into the encoder should make it work better than externally - but it's more complexity: more tuning in the encoder; more work in the decoder. Has anyone ever done it?

Things such as deblocking-filter and B-frames (framerate upconversion) have been integrated into codecs, even though they initially seem to have come from outside the codec. Reason seems to be that they had good PQ to bitrate/complexity ratios and they could do better inside the codec than outside.

I think that all sense indicates that if the source is progressive (not always true), then doing interlacing within the codec will give major benefits for image quality and possibly total complexity as opposed to doing it externally. Advanced deinterlacers do all kinds of "artificial intelligence" that they should not have to do given precise signalling on how the content was actually produced. Motion vectors could be jointly optimized for tracking motion and describing candidates for filling in lines, saving a lot of cycles and having the luxury of optimizing for the ground-truth in the encoder.


It might be that I/we are setting the wrong background for the discussion. 1080p50 is not generally the source, and if one made 1080p50 cameras, they would have worse noise-performance. If that is the case, then interlacing could be a reasonable technology in the camera to overcome sensor limitations. If that is the case, then it may be the case that deinterlacing in the camera to 1080p50 does not increase quality/bitrate sufficiently, but does increase complexity considerably. I dont know.

-k

[MfA]

look sharper on most TVs - even the previously common 768-line ones
It can look sharper in theory, but only if you allow aliasing ... if you remove all the aliasing you are essentially going to be halving the vertical resolution (which is why generally some aliasing is left and they reduce vertical bandwidth by ~70%).

[knutinh]

It can look sharper in theory, but only if you allow aliasing ... if you remove all the aliasing you are essentially going to be halving the vertical resolution (which is why generally some aliasing is left and they reduce vertical bandwidth by ~70%).
I guess that one way of looking at it - assuming perfect content-adaptive interlacers and/or deinterlacers (prefe
reably exchanging metadata) would be a system capable of 1920x540@50p <-> 1920x1080@25p and anything in-between on a spatial/temporal as-needed basis. When doing something like a seemingly perfect 1920x1080@50p linear pan, that would be based on (possibly sensible) assumptions about how scenes are captured, but still doing bad errors.

Superresolution systems depend on aliased input. I think that they overlap a lot with interlacing (at least in theory).

Do you think that it is possible to use correlation to estimate if the interlacer used "field integration mode" or "frame integration mode" (or possibly some 70% vertical filtering), and use that information to select between different modes of deinterlacer agressiveness?

-k

[2Bdecided]

It can look sharper in theory, but only if you allow aliasing ... if you remove all the aliasing you are essentially going to be halving the vertical resolution (which is why generally some aliasing is left and they reduce vertical bandwidth by ~70%).Well, if you have a 1366 x 768 display, then obviously a 1920 width source will look sharper than a 1280 width source in the horizontal direction. There can be no argument there.

Whether the 1080 line interlaced version or 720 line progressive version looks sharper depends on the factors you mention. At best, 1080 can look sharper (by as much as the 1080:720 ratio suggests), with dumb deinterlacing they're quite similar, but the 1080i version will visibly bob. It's rare for interlaced signals to be filtered to half the vertical resolution, so suggesting you'll get 540 vs 720 isn't realistic.

Cheers,
David.

[2Bdecided]

For your statements to be generally right, I think one would expect that compressing any original 1080p50 sequence at:
1)1080@50p, h264, X mbps
2)1080@50i, h264, X mbps
3)720@50p, h264, X mbps


Like this...

http://www.ebu.ch/CMSimages/en/tec_ebu_ibc2006_hand-out_hdvt_2_HQ_tcm6-46693.pdf

In 2006, the EBU (inc the IRT, SVT, etc) tried very hard to convince everyone that 1080i wasn't worth it. A year earlier, the IRT were doing demonstrations of this, intentionally using the worst MPEG-4 codec they could find wrt interlacing capability!

Yet when it came to launching HD across Europe, broadcasters chose 1080i. The reason they give is that with newer encoders and full HD displays, 1080i is the current sweet spot.

Part of the problem is probably that they don't have 1080p easily available as a delivery format yet.


It's easier to do the test at SD resolutions, and just as valid. If interlacing is useless, then 720x576p50 at a given bitrate should always look better than 720x576i50 at the same bitrate.

With x264, I think that might be true. But broadcasters are saying that the hardware encoders they have available don't give this result.

I haven't seen any published tests that match this - quite the opposite...
http://ip.hhi.de/imagecom_G1/assets/pdfs/ieee_sub_quality_hdtv_2008.pdf
...either the broadcasters are lying - or encoders have changed since that paper was written. Note: they create a 1080i50 signal using about the worst possible method in that paper.

Cheers,
David.

[Manao]

It's easier to do the test at SD resolutions, and just as valid.Not really. SD on a HD TV looks ugly, and on a CRT, you can't show 576p.

[knutinh]

Well, if you have a 1366 x 768 display, then obviously a 1920 width source will look sharper than a 1280 width source in the horizontal direction. There can be no argument there.

Agreed. Smart displays could even benefit from a little subpixel scaling.


Whether the 1080 line interlaced version or 720 line progressive version looks sharper depends on the factors you mention. At best, 1080 can look sharper (by as much as the 1080:720 ratio suggests), with dumb deinterlacing they're quite similar, but the 1080i version will visibly bob. It's rare for interlaced signals to be filtered to half the vertical resolution, so suggesting you'll get 540 vs 720 isn't realistic.

Cheers,
David.
The camera info that I found suggested that native 1080i capture will either be:
1)Non-filtered (at least electronically), meaning that you let through all aliasing allowed by the transfer function of optics and Optical Lowpass-filter.
2)Sensor line#1 and line#2 is averaged to produce line#1 of field#1. Line#2 and line#3 is averaged to produce line#1 of field#2. I believe this to be a vertical 2-tap boxcar pre-filter. It has a null at fs/2, lets through significant aliasing between fs/4 and fs/2, and attenuates some passband detail below fs/4.

For cases where interlacing is applied digitally on a progressive source, there should be many more options. Either tailor-made static filtering, or scene-adaptive filter cutoff. Do you know anthing about what the actually do?

For embedding 1080@24p inside 1080@60i (or 1080@25p inside 1080@50i), I think that they should employ no vertical filtering.

BTW, nice to see that hydrogenaudio-members are into video as well.

-k

[knutinh]

Like this...

http://www.ebu.ch/CMSimages/en/tec_e...tcm6-46693.pdf

I only found a single page describing a setup. Was there supposed to be any results?


http://ip.hhi.de/imagecom_G1/assets/pdfs/ieee_sub_quality_hdtv_2008.pdf

A good one. Thank you.


Yet when it came to launching HD across Europe, broadcasters chose 1080i. The reason they give is that with newer encoders and full HD displays, 1080i is the current sweet spot.

But they are not academics. If the market responds more positively to "1080" than "720", they will offer it, no matter if it is technically "better", wont they?

Part of the problem is probably that they don't have 1080p easily available as a delivery format yet.

Why is it a problem to use 720p?

It's easier to do the test at SD resolutions, and just as valid. If interlacing is useless, then 720x576p50 at a given bitrate should always look better than 720x576i50 at the same bitrate.

I think that you are right. By having a display that has far higher resolution than the content, we can effectively "factor it out".

It might be that tests at 576i/576p/384p should be carried out at larger distances/smaller displays to be representative of 1080i/1080p/720p.


I believe that the tv-industry is quite conservative. Where IT change equipment and mindset every 3 years, these guys tends to have 20 year cycles. They have invested heavily in editing equipment and interfaces that is limited to 1080i. The big manufacturers have an interest in differentiating themselves through superior deinterlacing. For cameras, there seems to be a potential advantage to do native interlaced capture. For 24p content, they have a working (sort of) channel using 60i/50i.

-k

[MfA]

It's rare for interlaced signals to be filtered to half the vertical resolution, so suggesting you'll get 540 vs 720 isn't realistic.
As I said, the bandwidth is usually reduced to around 70% ... so the effective resolution is about 70%.

That's just when things aren't moving though. When things move at a multiple of 1 pel per field in the vertical direction even the best deinterlacers being used at the moment are not going to be able to make something sharp out of the result ... the object becomes identical (ignoring the shift) in both fields, all the image data for half the lines in every frame is simply gone, interpolation will have to do. So in that case the effective resolution becomes 50% ... but it's aliased, so it's actually a little worse than 50%.

Which is why you really don't want to do high movement video with interlacing (sports mostly, since action movies are of course shot with flicker cam, making interlacing moot).

[knutinh]

As I said, the bandwidth is usually reduced to around 70% ... so the effective resolution is about 70%.

That's just when things aren't moving though. When things move at a multiple of 1 pel per field in the vertical direction even the best deinterlacers being used at the moment are not going to be able to make something sharp out of the result ... the object becomes identical (ignoring the shift) in both fields, all the image data for half the lines in every frame is simply gone, interpolation will have to do. So in that case the effective resolution becomes 50% ... but it's aliased, so it's actually a little worse than 50%.

Which is why you really don't want to do high movement video with interlacing (sports mostly, since action movies are of course shot with flicker cam, making interlacing moot).

I just remembered a review of Canon Hf20 consumer 1080@60i AVCHD camera:
http://www.camcorderinfo.com/content/Canon-Vixia-HF20-Camcorder-Review-36314/Motion-amp-Resolution-Performance.htm


The Canon HF20 has the best video resolution we've ever recorded on a consumer camcorder and its scores are comparable to some of the professional models we've tested (like the Sony HDR-FX1000 and Canon XL H1A). The camcorder measured an approximate video resolution of 800 line widths per picture height (lw/ph) horizontal and 900 lw/ph vertical.

That would seem to indicate that little to no vertical prefiltering is used for this particular camera, wouldnt it?

-k

[MfA]

Not exactly, because generally we don't use brickwall filters in video filtering ... just for instance a simple 1/4, 1/2, 1/4 filter will let such high frequency line patterns through attenuated. You can't exactly say it leaves resolution intact though.

PS. if you are going to ignore my argument of vertical resolution in the presence of vertical motion don't quote it ...

[Manao]

That would seem to indicate that little to no vertical prefiltering is used for this particular camera, wouldnt it?I wasn't able to find any equivalence between lw/ph and effective resolution in pixels. What I was able to find, however, was that the IPhone has a lw/hp of 866 horizontally and 897 vertically. Which makes me doubt the quality of this camera. Care to explain ?

[knutinh]

Not exactly, because generally we don't use brickwall filters in video filtering ... just for instance a simple 1/4, 1/2, 1/4 filter will let such high frequency line patterns through attenuated. You can't exactly say it leaves resolution intact though.

I was under the impression that such filtering was not possible in the sensor itself. If my sources are right that this sensor is native 60i, then it seems that electronic filtering will either be [1, 1]/2, or none. As for OLPF and optic flaws my knowledge is worse.

A tent filter will still have considerable attenuation above its cutoff frequency. If the measurement technique is good, then quoted line-pairs (or whatever) should be representative of the real frequency response, should it not?


PS. if you are going to ignore my argument of vertical resolution in the presence of vertical motion don't quote it ...

I am not ignoring it. But for resolution in static scenes it is irrelevant. The question is to what degree cameras employ static filtering (for combating issues with movement) that also degrade the resolution in non-moving scenes. The answer seems to be that both are possible, like I said a few posts ago:

To complicate matters further, different cameras construct the two video fields in different manners. In some cameras the even field corresponds to the even lines of pixels in the CCD chip, and the odd field to the odd lines of pixels in the CCD chip.
...
Slightly better are cameras which produce an average of the even lines and the preceding odd lines for the even field, and the odd lines and the preceding even lines for the odd field.

If you lookup that post you will see that I supply urls.

-k

[MfA]

No, it's not representative of the frequency response because in and of themselves you can't determine how fast the fall off (ie. how blurry vs ringy it is) is from a single data point.

I never said you didn't supply URLs ... hell the more the merrier, let me do one (http://www.ebu.ch/en/technical/trev/trev_301-editorial.html) too. It's from EBU which has been doubted in this thread, but even if you doubt them there is the zinger from Faroudja (yes, that one) ... “I am amazed that anybody would consider launching new services based on interlace. I have spent all of my life working on conversion from interlace to progressive. Now that I have sold my successful company, I can tell you the truth: interlace to progressive does not work!”.

[knutinh]

No, it's not representative of the frequency response because in and of themselves you can't determine how fast the fall off (ie. how blurry vs ringy it is) is from a single data point.

If a decimation/interpolation pass has a passband that is within +0dB/-3dB from DC to frequency Fc, then I would claim that that single datapoint says a lot about the system.

Ringing is first and foremost an issue in high-order filtering including negative coefficients, I think. Do you see much ringing in 2-3tap decimation/interpolation filters?

A system with flat and wide passband should be relatively unblurry. The degree of blurryness should be relatively well predicted from the "single datapoint" of passbandwidth?

there is the zinger from Faroudja (yes, that one) ... “I am amazed that anybody would consider launching new services based on interlace. I have spent all of my life working on conversion from interlace to progressive. Now that I have sold my successful company, I can tell you the truth: interlace to progressive does not work!”.
Hilarious. Like I said in my first post in this thread (before venturing off into nit-picking):
I tend to see interlacing as an analog 2:1 perceptually motivated compression method. I dont really see its purpose in this digital era, except for legacy purposes. If you want to trade motion/resolution/bandwidth, then use a lossy digital codec that does it intelligently..

-k

[knutinh]

I wasn't able to find any equivalence between lw/ph and effective resolution in pixels. What I was able to find, however, was that the IPhone has a lw/hp of 866 horizontally and 897 vertically. Which makes me doubt the quality of this camera. Care to explain ?

That camera was in fact one of the highest effective resolution consumer 1080 cams (it fails in other areas such as low light performance though).

It would seem that their "video sharpness" test does in fact factor in temporal aspects such as interlacing/deinterlacing, temporal lossy compression etc (my assumption was wrong. Always delightful to improve understanding):

Video Sharpness
The sharpness that a camcorder actually produces is rarely the same number that the manufacturer advertises. For instance, camcorders that output a 1920 x 1080 picture are not actually capturing one thousand nine-hundred and twenty horizontal lines of information. That's simply the size of the "container" that the camcorder outputs (also known as the resolution). In fact, there are lots of ways that manufacturers can play with the numbers, emphasizing capabilities of the lens, or the sensor, or something else. The simple fact is, you don't buy a sensor, and in most cases, you don't buy a lens. You buy a camcorder – a complete, pre-assembled camcorder, so that's how we test them.

We light a DSC Labs Multiburst chart at an even 3000 lux. The camcorder is stationed in a fixed position on a tripod. We aim the camcorder, aligning with the chart's 16:9 guideframes. We then pan the camcorder slowly left and right for about 30 seconds. Then we re-align the camcorder and tilt slowly up and down for about 30 seconds.

After the shooting is complete, we connect the camcorder to our HDTV using the camcorder's highest quality connection, typically either composite-out, S-video, component-out, or HDMI. We examine the playback footage, looking for the point at which the lines on the chart become indistinguishable.

The reason we test sharpness with the camcorder in motion, rather than a static shot, is simple. When was the last time you shot a video with nothing moving? The inherent nature of video is movement. This may not be the method manufacturers would prefer, but we think it makes the most sense.

The final score for this section is based on the horizontal and vertical sharpness as recorded in auto mode in the 60i frame rate. We may also examine the sharpness in other frame rates, but it does not factor into the score.


http://www.imatest.com/docs/sharpness.html


...
The use of Picture Height gives a slight advantage to compact digital cameras, which have an aspect ratio (width:height) of 4:3, compared to 3:2 for digital SLRs. Compact digital cameras have slightly more vertical pixels for a given number of total pixels.
http://www.imatest.com/docs/images/MTF_EOS-40D_EF1785_26mm_f56.gif

[2Bdecided]

...That's just when things aren't moving though. When things move at a multiple of 1 pel per field in the vertical direction even the best deinterlacers being used at the moment are not going to be able to make something sharp out of the result ... "even the best deinterlacers" ever can't+won't overcome the fundamental high spatial frequency / temporal confusion inherent in an interlaced signal.

Lots of movement and/or lots of fine detail isn't a fundamental problem - the fundamental unrecoverable problem is the specific combination of fine detail and specific movement, as you illustrated.

Where there's movement + fine detail that is recoverable, it's often argued that it doesn't matter so much if the display can't recover it because we're less sensitive to detail when the image is moving. It's not true for eye-tracked motion however - but then most modern displays are already a disaster for eye-tracked motion anyway, so it matters less.

Cheers,
David.

[MfA]

If a decimation/interpolation pass has a passband that is within +0dB/-3dB from DC to frequency Fc
How could it be any different? The -3dB point is the definition of the cut off frequency. We don't know the cut off frequency though, that's not what they are measuring (the lines will still be visible long after that). Higher order flicker filters do exist and even the tent and averaging filters differ by an order, so you really need two datapoints.

Sorry, I got you and 2B mixed up ... thought you were arguing in favour of making interlacing a standard in a time period when progressive displays already ruled the roost (the US had a decent excuse since they started so much earlier).