Interframe Long-Gop Progressive

[Essor Nougier]

How does Long Gop interframe HDV compression (compressing frames based on similarities and differences with one another), which only has one full frame every 15 frames, work with P shooting? Progressive frames.

How does long gop work then? I can see how it encodes split frames, but full frames?

[Giovanni Speranza]

It depends on the camera.
AFAIK HDV has many flavours.
And processing before encoding is a key factor, i.e. it seems that the f mode of the Canon XH A1, which interpolates the fields to create a progressive frame before encoding results in a better compression because there is less information to encode.
So i think that progressive HDV is better than interlaced because there is less data to compress, or say less image variations.

There are so many factors that influence the image quality...

[Essor Nougier]

Yes, but theoretically, any camera that has a native progressive sensor and records to HDD or solid state memory, even if using long gop compression, should record only full frames, not a full frame every 15 frames.

[David Heath]

Originally Posted by Essor Nougier

Yes, but theoretically, any camera that has a native progressive sensor ........ should record only full frames, not a full frame every 15 frames.

Not really. Think of it as a three stage process - initial acquisition, recording, and final output. If it captures true progressive frames, and outputs the same, what does it matter what happens in between?

And the theory of long GOP compression relies on taking advantage of similarities between successive frames to gain an advantage by only recording differences between frames, on the assumption that they take up less bits than the frames themselves. In theory, ALL the frames can then be reconstructed perfectly for output, and the user is unaware of what has happened in between. Generally, progressive systems lend themselves much better to long GOP compression than interlace, and is one of the arguments for using p rather than i, certainly for broadcast transmission.

In practice, things are never perfect, and how far off depends on the bitrate compromise - if there are too many changes between frames (eg excessive motion) "blocking" starts to appear. This can be very noticeable in the case of some digital transmissions (where way too small bitrates are used) or so slight that any artifacts are masked by the motion anyway. Very, very approximately, if used to gain a coding efficency of 2-3x relative to a comparable intraframe system, artifacting should be negligible. Go much higher and expect problems with high motion scenes.

[Chris Hurd]

Originally Posted by Giovanni Speranza

it seems that the f mode of the Canon XH A1, which interpolates the fields to create a progressive frame before encoding results in a better compression because there is less information to encode.

Sorry but that is technically incorrect. The Canon XL and XH series Frame mode does not "interpolate the fields." Instead it is a full-frame capture. It does not carry substantially less information than normal interlace.

So i think that progressive HDV is better than interlaced because there is less data to compress...

Also incorrect. 1080i and 720p carry the exact same amount of data.

[Giovanni Speranza]

This is a good news. I thought that the Canon f was a fake progressive (a really good looking one). Thanks.

[Daniel Epstein]

Originally Posted by Chris Hurd

Sorry but that is technically incorrect. The Canon XL and XH series Frame mode does not "interpolate the fields." Instead it is a full-frame capture. It does not carry substantially less information than normal interlace.

Also incorrect. 1080i and 720p carry the exact same amount of data.

Hey Chris Technically there is a difference between the Data Rates of 1080i and 720P according to what I have Read.

"The HDV video format supports MPEG-2 compressed video (MP@H-14), at two 16:9 resolutions:
- 720p (1280 x 720, progressive), at approximately 19 Mbps data rate
- 1080i (1440 x 1080, interlaced), at approximately 25 Mbps data rate"

http://www.manifest-tech.com/media_pc/hdv_jvc.htm

This accounts for some of the difference in GOP lengths

[Chris Hurd]

D'oh. Right you are, Daniel -- thanks for the clarification.

[Adam Letch]

is it outputting 1080i at 25mbps over firewire?

I thought they stuck to their 19.7mbps standard?

it would be interesting if it were the case, I know the hdtv standard is 19.7mpbs, but always thought that they should squeeze as much info in their codec as possible and use the full 25mpbs of the hdv standard, but keep their 6GOP structure.

[John Wyatt]

Can someone confirm for me something I read which said there was a different GoP range for NTSC and PAL HDV? I can't remember the source but noted down that PAL GoP was supposed to be 12 frames and NTSC was 15. If PAL does have a shorter GoP does this provide an advantage for PAL recording?

The fact that PAL presents less images per second than NTSC makes me wonder if PAL HDV might slightly better be able to handle more content changes in a frame before artefacting? If the data rate is the same for both (25 mbps), then that datarate only has to make 50 fields a second with PAL, but with NTSC the same 25 mbps has to be shared between 60 fields (so there's less data for each field?). Does this have a bearing on image quality, and if so is it insignificant or of practical value?

And to return to the subject of the thread, since 1080 PsF progressive is recorded to tape as interlaced, what effect do the above points have on progressive quality regarding NTSC versus PAL?

[Paul Kellett]

You could've asked me that John.

My cam is mpeg-2 long gop,12 frames.

Paul.