Picture per field encoding

[Alex-Kid]

That's it. I need an explanation about picture per field encoding.

I'm searching a video encoder that satisfies my quality needs of converting DV footage to MPEG-2 for DVD authoring. My man Google found information about picture per field encoding (GSpot) and, as far as i know, consists of encoding individual fields as frames, with their own bitrate and quantization, and then "weave" them. I'd like to know if i'm right or wrong, if that's better or worse than common interlace encoding and the encoding applications that use this method.

Thanks in advance, saludos desde Chile

By ALEX-KID

[manono]

Hi-
...consists of encoding individual fields as frames, with their own bitrate and quantization, and then "weave" them.
If the picture structure is field, then the fields are encoded and then stored on the DVD separately. They're not woven back together. There are flags set that tell how and in what order they're to be displayed.

Although I have some DVDs where the picture structure is field, they are really horrible DVDs with other things wrong with them. I've read that of all the software encoders in commmon usage, only Procoder is supposed to be able to do what you're asking, but I've never read of anyone actually succeeding at it.

There's a small advantage to encoding interlaced material (and particularly DV footage) using a field pic structure, in that I-Frames are fields instead of full frames. Therefore the I-Frames are smaller than usual, and theoretically the resulting DVD quality might be a bit higher for the same file size when compared to using a frame picture structure. But it's an academic question as I don't think you'll ever find a way to do any field pic structure encoding. I'll bow to my betters that may wish to elaborate on or contradict anything I've said.

[scharfis_brain]

procoder supports field encoding, like manono already said.

But it is only effective if the source video is
1) uncompessed / lossless
2) not already encoded frame-based (like DV does)

so there is no gain in field-based encoding of DV-Video.
I had a gain by fieldbased encoding, when I did some 50fps screencapture of some 3D-demos to DVD...

[Alex-Kid]

I tested Procoder this weekend and I must say I got the best results of all encoders tested (TMPGEnc and HCEnc, imho). My footage is about Petrohué falls (Saltos del Petrohué) and the water flowing looks very blocky and blurry in the encoded file, but this artifacts are reduced when using Procoder.

I also tested one of the sequences where water is flowing to determine what kind of picture structure is the best. The result (based on SSIM) says is best using field picture structure. But when encoding non water flowing sequence (a road in the middle of the forest) field picture structure is the best too. So, my conclusion is, for DV footage, always use field picture structure...

Another thing I found is that Procoder seems to have a more stable quantization. TMPGEnc and HCEnc showed high differences among the quantization factors in every second, compared with Procoder.

¿SSIM for Procoder, HCEnc and TMPGEnc? Procoder is lowest, because its SSIM had very low value at the beginning and in the end of video stream, but when measuring 50% of the frames in the middle, Procoder is highest. I don't know why this happens.

¿Any comments?

Saludos

By ALEX-KID

[manono]

¿Any comments?
1) We don't know any of the settings for any of the encoders, so, personally, I'd take those observations with a grain of salt.
2) I can believe that Procoder might do a better job with a hard to compress DV source, at least partly because it has a natural smoothing filter built in that can't be turned off. I don't know how much of your supposed better quality results from the filtering it does, or from the fact that you used field-based encoding. My hunch is that you'd get very similar results using Procoder and frame based encoding. For a clean, detailed, and easy to compress progressive source, I'd surely prefer to use CCE or HCEnc.
3) Have you authored, burned to disc, and played the field-based encoding in your standalone DVD player? I've read several reports that standalones won't play Procoder encoded field picture structure DVDs.

[Alex-Kid]

Thank you manono, I knew there are some things i should consider. I'll do some other tests (PSNR and blurring) to confirm what you are saying. I will also watch the field based encoding in my standalone DVD player.

The settings used were: 2-pass VBR, average bitrate=6300, min bitrate=0, max bitrate=9800, interlaced BFF, scene change detection on, closed gop, MPEG matrix (no setting available for Procoder). For each encoder
HCEnc: high quality, lumgain=3, autogop=18 frames (NTSC).
TMPGEnc: highest quality, GOP size=18
Procoder: highest quality, VBV=224, field based structure, GOP size=auto.

However, I'm very surprised with Procoder quantization. As far as I know, quantization value is inverse to quality, so lowest quantization means better quality. Procoder has lowest quantization in about 70% of the water flowing video stream (remember, my video is DV).

When I get home, I'll paste some figures.

Saludos

By ALEX-KID

[scharfis_brain]

you also may try a GOP without B-Frames in difficult scenes.

[manono]

However, I'm very surprised with Procoder quantization. As far as I know, quantization value is inverse to quality, so lowest quantization means better quality. Procoder has lowest quantization in about 70% of the water flowing video stream (remember, my video is DV).
But only if everything else is the same. And it isn't. The smoothing that Procoder does also lowers the average quant. Does that mean it has the best encodes? Maybe, maybe not.

[scharfis_brain]

I am very pleased with the output of procoder.
Other encoders I tried were too much concentrated on maintaining the edges. This caused severe Ringing, BLocking and GOP-Pumping.
ProCoder avoids this by blurring the image slightly, if needed.
With high bitrates Procoder hasn't blurred my videos at all...

[kolak]

you also may try a GOP without B-Frames in difficult scenes.

GOPs without B frames and also field encoding mode makes Mpeg2 stream not working on some players (maybe 5%).


Andrew

[scharfis_brain]

Have you got any documents regarding this issue?

[45tripp]

As far as I know, quantization value is inverse to quality, so lowest quantization means better quality.
you may also run into linear vs non-linear scaling,
in which case you can't just read the lower quant scale numbers.

I am very pleased with the output of procoder.
Other encoders I tried were too much concentrated on maintaining the edges. This caused severe Ringing, BLocking and GOP-Pumping.
ProCoder avoids this by blurring the image slightly, if needed.
With high bitrates Procoder hasn't blurred my videos at all...

I agree with scharfis.
I hate ringing.

Have you got any documents regarding this issue?

probably personal experience/estimation

GOPs without B frames and also field encoding mode makes Mpeg2 stream not working on some players (maybe 5%).

is that either/or?
or the combination of both?

[kolak]

Have you got any documents regarding this issue?

No, spec allows for this (but this is only a spec:) ). It's more personal (and not only) experience.

Field encoding or GOPs without B frames causes problems on some players- it doesn't have to be in combination.

Cinemacraft Xtream has special option to avoid GOPs without B frames.

Andrew

[Alex-Kid]

Hello, I'm back :)

The figures I promised come into XLS format and contains graphs, sample frames and tables of PSNR and SSIM analysis for HCEnc 0.22, TMPGEnc 2.5 and Canopus Procoder 3 (frame based, field based and automatic settings). PSNR analysis shows very low values for Canopus (all settings), which would explain slight blur applied by the encoder; SSIM analysis shows a little higher values for Canopus (all settings), which would demonstrate that the amount of blur appied doesn't affect human visual perception.

http://es.geocities.com/alexkidch/analysis.xls

I'm satisfied with quality of Canopus 3: less or weak blocking and almost no ringing in high motion scenes. For all other scenes I used HCEnc.

Saludos

By ALEX-KID

[45tripp]

http://es.geocities.com/alexkidch/analysis.xls



[foreign language text removed]

[Alex-Kid]

I don't know what happen with the direct link. When the page loads, it shows another url (es.share.geocities). What you can do is erase ".share" and the file will finally load.

Saludos

By ALEX-KID