Log in

View Full Version : interlacing discussion


video
17th March 2005, 16:23
Originally posted by Manao
video :Of course, if you want to play back your video on a progressive screen, you ought to deinterlace it prior to encoding.

But many people want to play their interlaced video on their TV. In that case, interlaced ought to be preserved, because simply deinterlace it will make the motion slightly jerky, and restore it to 60 fps progressive will make the encoding less efficient than keeping it interlaced, without bringing any quality. Also, it will stress some more the decoder. [/B]
Afaik there is no standalone AVC capable palyer, and even full ASP support is problematic (3WP GMC, QPEL..). So playing an AVC video, means playing the video on a computer today. Interlace is not suitable for computer displays.
In case of interlaced material mpeg predictions are happening on a field basis, not on a frame basis, so there are cases when an interpolated 50/60 fps progressive material is more compressable than its interlaced master. Of course, playing back 50/60 fps progressive material may double the necessary cpu power compared to 25/30 fps progressive.

Manao
17th March 2005, 16:29
so there are cases when an interpolated 50/60 fps progressive material is more compressable than its interlaced master.No because, the interlaced master contains half the data of the progressive one.
Afaik there is no standalone AVC capable palyer,Luckyly, they'll come soon enough. A lot of HDTV will also be encoded in AVC / H264, and most of these content will (alas) be interlaced, so standalone will have to support interlaced encoding, and it may happen that we have to handle interlaced streams on our computers.

video
17th March 2005, 17:24
Originally posted by Manao
No because, the interlaced master contains half the data of the progressive one.

No, You are wrong. The interpolated material contains exactly the same amount of information as the interlaced one. If you are playing back a CD on a player using 256 times oversampling, you are hearing no more information, compared to the players not utilising oversampling at all. (The only thing, what makes the difference is that, a delta-sigma modulator can be manufactured much easier in good quality, and so the analog anti imaging filter) The same applies for the interpolated video, and during frequency-temporal domain analysis your AVC encoder will figure that out.

stephanV
17th March 2005, 17:41
Originally posted by video
No, You are wrong. The interpolated material contains exactly the same amount of information as the interlaced one. If you are playing back a CD on a player using 256 times oversampling, you are hearing no more information, compared to the players not utilising oversampling at all. (The only thing, what makes the difference is that, a delta-sigma modulator can be manufactured much easier in good quality, and so the analog anti imaging filter) The same applies for the interpolated video, and during frequency-temporal domain analysis your AVC encoder will figure that out.

not really true, an interpolator (in case of interlaced video a bobber) will actually create new data. Since an encoder cannot differentiate between real data or interpolated data it is very likely the video will be less compressible.

video
17th March 2005, 17:52
Originally posted by stephanV
not really true, an interpolator (in case of interlaced video a bobber) will actually create new data. Since an encoder cannot differentiate between real data or interpolated data it is very likely the video will be less compressible.
a bobber is hard to see as a real interpolator. I'm talking about real interpolator using mc. The same stuff what you can find in the so called '100Hz' Tvs. bobber works only in the spatial domain. a real interpolator operates in the temporal-spatial domain, ie. processes several frames before and after the current.

stephanV
17th March 2005, 17:58
then there is still interpolated (=extra) data. there are also motion based (temporal-spatial) bobbers BTW.

video
17th March 2005, 18:50
Originally posted by stephanV
then there is still interpolated (=extra) data. there are also motion based (temporal-spatial) bobbers BTW.

oh yeah. as you moving towards from cheating to real decoding - interlacing is nothing else, but a video codec compressing the bandwith to its half - your mpeg4 encoder will honor your efforts with better compression efficiency :)

stephanV
17th March 2005, 19:14
im not sure i understand what you mean now...

*.mp4 guy
17th March 2005, 19:46
Interlacing was originaly used in Tv. as a CODEC of sorts. It is now Very obsolete, so much so that we would be better off without it. This is because when you interlace something you loose data (most of the time). when you deinterlace something you lose data (more or less all the time) most modern video codecs work more efficiently on well deinterlaced data because it is easier to compress something that is not already compressed with a different scheme. for instance use levels (Vdub filter) or something to make a video pure black and white (as in 2bits) it wont compress nearly as efficiently with modern codecs as you would expect because that is not the type of data they work best on.

temporance
17th March 2005, 20:40
Sounds like people are confusing extra data with extra information. You get extra data by interpolating, sure, but there is no more information there.

stephanV
17th March 2005, 21:13
of course, and if you upsize a video with a factor 2 you can still compress it with the same bitrate and get the same quality as you would with the original resolution... or not :)

*.mp4 guy
18th March 2005, 00:31
Originally posted by stephanV
of course, and if you upsize a video with a factor 2 you can still compress it with the same bitrate and get the same quality as you would with the original resolution... or not :)

That is the point precisely, CODECs works best on data when it is already organized in certain ways.

stephanV
18th March 2005, 08:56
and by that you mean?

Manao
18th March 2005, 09:11
video : look at the problem like that : an interlaced video @ 30 fps is roughly seen by the encoder as a progressive video @ 60 fps, with half the resolution. What you say is that a progressive video @ 60 fps with the full resolution will be encoded better than the progressive one at half resolution. It can't be.

video
18th March 2005, 11:06
Originally posted by Manao
video : look at the problem like that : an interlaced video @ 30 fps is roughly seen by the encoder as a progressive video @ 60 fps, with half the resolution. What you say is that a progressive video @ 60 fps with the full resolution will be encoded better than the progressive one at half resolution. It can't be.

No. I think You have misunderstood what I've said. I was talked about interpolated 60 fps progressive, not just plain 60 fps progrssive.
Anyway
a) mpeg intra coding is more efficient with larger image sizes
b) doing P or B frames on the interlaced (field) basis is far less efficient than doing it for real, progressive material. On the interlaced basis the encoder will detect a lots of FALSE movement.

Teegedeck
18th March 2005, 11:18
How does all this relate to the topic of this thread?

Manao
18th March 2005, 11:21
> a) mpeg intra coding is more efficient with larger image sizes

OK, so first, what's that have to do with the problem at hand ? Second, it's plain false. AVC intra coding efficiency, in comparison to JPEG2000, decreases with the picture size.

> b) doing P or B frames on the interlaced (field) basis is far less efficient than doing it for real

What to you think is interlaced support in an encoder. If it's enabled, the encoder doesn't see a 30 fps interlaced video, but a 60 fps progressive one, at half the resolution. So it's _as_ efficient as coding a 60 fps progressive at half the resolution.

Hence :

> On the interlaced basis the encoder will detect a lots of FALSE movement.

Is not an issue anymore.

Finally

> I was talked about interpolated 60 fps progressive

Yeah, well such interpolations are rather hard to achieve. Mainly, they either suffer of blurring ( so indeed, they become easier to encode than full natural progressive 60 fps, though they'll still be harder to encode than half-res progressive 60 fps or 30 fps interlaced ), or they suffer from artifacts ( aliasing & co ) which make it a lot harder to encode.

Edit : Teegedeck, indeed you're right, can you move that discussion to another thread ?

Teegedeck
18th March 2005, 11:27
Nope, I can't; but I PM'd bond.

bond
18th March 2005, 11:34
Originally posted by Teegedeck
Nope, I can't; but I PM'd bond. you wish, we play :D