Is 704x480 with a 16:9 DAR valid for DVD?

[jagabo]

What the **** are you all talking about???

MPEG header info.

[intracube]

So when they regeared for anamorphic wide later in the 90's, why did they extend out to 720*?
The 2.35 :1 DVD movie authors took liberty with movie resolution as a creative decision thus they could say cropped 704 was the movie with 720 adding extra width.

The film standard '2.35' has changed over the decades - it's now 2.39:1 (approx.)

Most of my 2.39 (anamorphic 16:9) DVDs from the late 90s only use 702x576 for the active picture area (and no sequence_display_extension to be found). They were possibly sourced from broadcast Digibeta/BetacamSP.

2.39 DVDs that use the full 720 width are usually 'Digitally Remastered' and/or after 2001.

One would vertically crop 1920x1080 to 1920x1056 before downsize to 720x480 or 720x576. The center 704x480 or 704x576 would represent 16x9 at approximately correct aspect ratio.

I don't like the sound of that, tbh. Seems a messy solution that only prevents those displaying 720x480/720x576 from seeing 8px tiny bars - they would also have to be watching 720x480/720x576 at a ratio of 16.36:1...

To most people, a few percent inaccuracy in the proportions of the image should be nothing to worry about. But it could be a problem if video is routed through different equipment/software, each handling things differently. The cumulative effect might be pretty noticeable - so it's a good to get a clear idea on what's going on.

Does the stream also have picture display extensions?

I don't think so - can't find '00 01 B5 7something'. The disc played in a standalone player (set to 4:3 Pan Scan) and gives a hard letterboxed 16:9 image on a 4:3 TV.

Taking a step back, for a moment. Aren't a lot of the flags in the mpeg stream overridden/made redundant by the DVD authoring software?

I can create a MPEG-PS stream with the aspect ratio of 4:3 using ffmpeg. With dvdauthor, that can be over-ridden to 16:9 in its xml config file. A DVD player will display it as 16:9, even if the headers in the original stream are set to 4:3.

***
Wikipedia states that VOBs are 'based on the MPEG program stream format, but with additional limitations' and 'VOB files are a very strict subset of the MPEG program stream standard'.

So, while options like sequence_display_extension exist in the more general iso13818-2, they might not be be mandatory in the DVD spec (DVD-Video Book) or could even be ignored if they happen to be in the stream.
***

[jagabo]

Well, this is only one data point but...

My Philips 5990 DVD player, upscaling to 1080p60, shows every pixel of a 720x480 16:9 frame, filling the 16:9 HDTV screen (Samsung LNT 4665 in "Just Scan" mode). Playing a 704x480 16:9 frame every pixel was visible but there were small black borders at the left and right edges of the screen. 4:3 showed similar results, th 704x480 frame was a little smaller on the screen than the 720x480. In other words, the Philips 5990 treats 720x480 and 704x480 as the same pixel aspect ratio (32:27) and 720x480 is required to fill the 16:9 screen. The MPG encodings were done with HcEnc, no sequence_display_extension according to GSpot, authored with Ulead DVD MovieFactory 4 (not re-encoded).

[Anonymous344]

That is how every player that I have ever had has displayed DVDs too. Both 704x480 DVDs and 720x480 DVDs with blacks pixels on the left and right have looked like that on exact (or just) scan. That is true of both commercial DVDs and DVDs that I have encoded with HcEnc.

[intracube]

720x480 DVDs with blacks pixels on the left and right have looked like that on exact (or just) scan. That is true of both commercial DVDs and DVDs that I have encoded with HcEnc.

These images are from Goldeneye (R2 DVD) 720x576 with 702x576 active area:

The form has put them in the wrong order. Example 1 (bottom) shows the bars to the left/right. Example 2 (middle) with the red square overlayed shows the gun barrel is distorted. Example 3 has been cropped to 702 wide before being displayed at 16:9, and looks accurate. I don't think the difference in line standard matters for this test - it looks like some DVDs at least aren't going to be displayed correctly with your setup.

[Anonymous344]

it looks like some DVDs at least aren't going to be displayed correctly with your setup.

I don't understand what you mean at all. Why are you scaling to 640x360? As I mentioned before, DVDs have always looked like this on every player I have ever owned, and on every TV I have ever owned that has an exact\just scan mode. Of course, I don't ever watch a DVD on exact scan. I always watch using 16:9 (wide) mode.

[jagabo]

I don't ever watch a DVD on exact scan. I always watch using 16:9 (wide) mode.

Why?

[Anonymous344]

I don't ever watch a DVD on exact scan. I always watch using 16:9 (wide) mode.

Why?

I thought DVDs were supposed to be authored with overscan in mind. 1:1 doesn't overscan, so as far as I was aware, watching a DVD on 1:1 would not yield the correct aspect ratio. I only watch Blu-rays/HD DVDs on exact scan.

EDIT: Well, perhaps it is more accurate to say nominal analogue blanking rather than overscan. What I mean is that the eight pixels on either side, whether black pixels or not, are not meant to be seen. When I view DVDs on exact scan, there is always evidence of junk pixels on either side that are not meant to be seen.

[intracube]

I don't understand what you mean at all. Why are you scaling to 640x360?

The 640x360 image represents a TV - 640x360 chosen because it's exactly 16:9, (for people viewing this thread on most computers, with square pixels) and because it's not too big for people viewing the forum with small computer screens.

As I mentioned before, DVDs have always looked like this on every player I have ever owned, and on every TV I have ever owned that has an exact\just scan mode.

Apologies if I've not understood - I thought you and jagabo were saying that the full DVD image was displayed exactly 'as is' on your 16:9 sets (including small black bars to the left and right on some discs). My examples were to show that in the case of 'Goldeneye', the gun barrel is not a circle unless the image width is cropped to 704/702 before being displayed as a 16:9 image.

Of course, I don't ever watch a DVD on exact scan. I always watch using 16:9 (wide) mode.

Sorry, I didn't realize this. Though I'm not sure how your TV formats the signal '16:9 wide mode'.

EDIT: Well, perhaps it is more accurate to say nominal analogue blanking rather than overscan. What I mean is that the eight pixels on either side, whether black pixels or not, are not meant to be seen. When I view DVDs on exact scan, there is always evidence of junk pixels on either side that are not meant to be seen.

Ah, I understand. How you have you TV configured normally for DVDs (cropping the sides) is exactly what I thought should happen.
But, I have other DVDs that are formatted differently to Goldeneye - and cropping the sides off them before displaying as 16:9 would make a circle end up too wide....

[edDV]

I still think this move to 720 by the Movie DVD authors was a cheat to narrow letterbox bars at the expense of a 2.3% side crop when viewed analog or digital at 704x480 or 702x576.

[Anonymous344]

Sorry, I didn't realize this. Though I'm not sure how your TV formats the signal '16:9 wide mode'.

No worries! I understand about the 640 by 360 images now. Sorry, mate; I've had a long day, and I'm not quite thinking straight!

To clarify, I don't watch a DVD with the pillarboxing on the left and right visible. It is my understanding that HDTV's have an exact (just) scan mode for Blu-ray and other HD material, not for DVD. As I understand it, upscaling players should upscale DVDs to 1964x1080, so they still need to be overscanned.

As to how my TV formats the signal, I understood that TVs traditionally overscan to different degress anyway. I put Wide in brackets as some manufacturers call the mode 16:9 and some call it Wide, just as some manufacturers call 1:1 Just Scan and some call it Exact Scan.

I still think this move to 720 by the Movie DVD authors was a cheat to narrow letterbox bars at the expense of a 2.3% side crop when viewed analog or digital at 704x480 or 702x576.

Interesting! What would be the motivation for that? Do you mean that some people don't like big letterbox bars, so this was a way to narrow them?

[jagabo]

perhaps it is more accurate to say nominal analogue blanking rather than overscan. What I mean is that the eight pixels on either side, whether black pixels or not, are not meant to be seen. When I view DVDs on exact scan, there is always evidence of junk pixels on either side that are not meant to be seen.

If they're black, who cares? And I've seen many DVDs where the picture goes all the way to the edge.

[edDV]

I still think this move to 720 by the Movie DVD authors was a cheat to narrow letterbox bars at the expense of a 2.3% side crop when viewed analog or digital at 704x480 or 702x576.

Interesting! What would be the motivation for that? Do you mean that some people don't like big letterbox bars. so this was a way to narrow them?

Yes.

Keep in mind this was also done for the DV camcorder format for both 4:3 and 16:9 wide. In the case of 4:3, the PAR was the same as DVD which means 4:3 is represented by the center 704x480 or 702x576. Same goes for wide. The 16 extra pixels represent additional width for both 4:3 and 16:9.

[Anonymous344]

[If they're black, who cares? And I've seen many DVDs where the picture goes all the way to the edge.

They are not always black. I have seen a number of transfers that have an ugly purple or green line visible at the far edge that is obviously not meant to be seen. The junk pixels at the side can be black pixels or they can be picture information from the transfer, but either way I don't believe that they are meant to be seen or the AR will be slightly incorrect.

In fact, it is rare for the black pixels in the pillarboxing to be exactly eight pixels; but even if, for example, there are five black pixels at one edge, I still think three further pixels that cut into what looks like the picture are not meant to be seen. (It is also common for there to be black pixels visible at one edge, but not the other.)

[jagabo]

[If they're black, who cares? And I've seen many DVDs where the picture goes all the way to the edge.

They are not always black. I have seen a number of transfers that have an ugly purple or green line visible at the far edge that is obviously not meant to be seen. The junk pixels at the side can be black pixels or they can be picture information from the transfer, but either way I don't believe that they are meant to be seen

If they're picture information they're meant to be seen. I don't think I've ever seen a commercial DVD with anything other than black at the edges.

[Anonymous344]

If they're picture information they're meant to be seen.

I don't understand. I thought that the picture that is meant to be seen, even in 720x480 frame, is contained in the core 704x480 area and that the eight pixels on either side are junk. If so, then whether or not the junk pixels are black or material from the transfer shouldn't matter. If all picture information is meant to be seen, then I don't see why some studios add black pixels to, for example, the left and not the right.

[jagabo]

I don't understand. I thought that the picture that is meant to be seen, even in 720x480 frame, is contained in the core 704x480 area and that the eight pixels on either side are junk. If so, then whether or not the junk pixels are black or material from the transfer shouldn't matter. If all picture information is meant to be seen, then I don't see why some studios add black pixels to, for example, the left and not the right.

So you'd rather lose 2 or 3 percent of the picture at each edge (typical overscan on a 1080p display) than see a half percent of black bars at the left and right? 16:9 mode on a Samsung LCD is just a zoom (to simulate overscan) of the Just Scan image.

[Anonymous344]

So you'd rather lose 2 or 3 percent of the picture at each edge (typical overscan on a 1080p display) than see a half percent of black bars at the left and right? 16:9 mode on a Samsung LCD is just a zoom (to simulate overscan) of the Just Scan image.

I would yes, because as I understand it, the image that is meant to be seen is in the 704x480 core anyway, so I am not losing anything that is meant to be seen. I use the 16:9 mode for DVD precisely because it simulates overscan, because it is my understanding that DVDs, if authored correctly, take overscan into account. I even thought that the 16:9 mode was there for that reason. I always switch to just scan when I watch a BD and switch back for DVD.

[intracube]

If they're picture information they're meant to be seen.

But seen at what aspect ratio? Should the 720x480 image be scaled to 16:9 or to 16.3632:9? No consensus was reached over the use of sequence_display_extension.

I don't think I've ever seen a commercial DVD with anything other than black at the edges.

The House DVD that I originally mentioned doesn't have any black. Neither does the Back To The Future trilogy (both R2 widescreen and R1 fullscreen) Amadeus (R2 widescreen - *almost the whole width), and many more that I've been checking.

[Anonymous344]

Intracube, there are many PAL DVDs like that. In my experience it is more common for PAL DVDs not to have visible pillarboxing on the left and right than it is for NTSC. I have never been able to make up my mind whether this is because:

1. these transfers do not take nominal analogue blanking into account because of studio error;
2. HDTVs are more commonly found at present in Europe than in the US (this is speculation, on my part), so studios assume that people will now be using a Just Scan mode on their televisions and so ignore nominal analogue blanking when authoring PAL DVDs
3. the transfer does take nominal analogue blanking into account but so much of the junk is material from the transfer that you have you to really zoom in to tell that the transfer doesn't meet the edge. (It rarely fully meets the edge if you zoom in using Photoshop.)

I can even give you an example of a transfer that has pillarboxing in the NTSC version, but not the PAL version. Star Trek II: the Wrath of Khan: Director's Cut. (I own both.)

[edDV]

If they're picture information they're meant to be seen.

I don't understand. I thought that the picture that is meant to be seen, even in 720x480 frame, is contained in the core 704x480 area and that the eight pixels on either side are junk. If so, then whether or not the junk pixels are black or material from the transfer shouldn't matter. If all picture information is meant to be seen, then I don't see why some studios add black pixels to, for example, the left and not the right.

The center 704x480 or 702x576 are what is seen after analog conversion or ATSC/DVB broadcast. Also most HDTV sets crop to 704/702 before upscale. But in "Just Scan" mode or on a computer (editors or players) the full 720 may be shown especially for anamorphic DVD or wide DV source. Most computer edit programs default to 720 rather than cropped 704. All my TV captures show the 8 pixel blanking edges.

It would be interesting to make a test file with bright Yellow 8 pixel side bars and test in various players and displays.

[Anonymous344]

To give an example of what I mean, the R1 North by Northwest DVD has a green discoloration running down the edge of the transfer. Now to me it seems that you either accept that this meant to be seen and that the studio made a grave error in quality control, or the edge of that image is junk (whether it looks like picture or not) and is meant to be overscanned.

http://www.dvdbeaver.com/film/DVDCompare5/northbynorthwest.htm

There are other DVDs with ugly lines this that I think aren't meant to be seen, and never would have been in the old days before TVs with Just Scan modes.

[jagabo]

I use the 16:9 mode for DVD precisely because it simulates overscan, because it is my understanding that DVDs, if authored correctly, take overscan into account. I even thought that the 16:9 mode was there for that reason. I always switch to just scan when I watch a BD and switch back for DVD.

You lose much more that 8 pixels at the left and right edges by using 16:9. I just measured on my Samsung. Using 16:9 crops about 24 pixels at the left and right edges, and 16 pixels at the top and bottom. The aspect ratio was the same in Just Scan and 16:9.

Obviously, this is a matter of personal preference. There's usually nothing important at the extreme edges of the frame. I usually use Just Scan -- unless there's something annoying like a green bar (I've only seen this on cable TV), tape head switching noise at the bottom of the frame, or closed caption encoding at the top of the frame.

Just in case it wasn't clear in my earlier post: My Philips 5990 DVD player upscales the full 720x480 frame to 1920x1080 (16:9).

[intracube]

Jeff, the answer might well be all three of the possibilities you state - depending on where the DVD is authored, who does the authoring, and the original source of the film (Digibeta/BetacamSP/HDCamSR/etc and how the film was originally transferred).

For now, at least when authoring a DVD:
- 704x480/704x576 seems a valid encoded frame size for 16:9 DVDs (confirmed by Shon3i).
- 704x480/702x576 are the correct sizes to pre-scale to if you have *actual* 4:3 or 16:9 footage.
- Using the full 720x480/720x576 frame *might* be allowed, but only if you've got slightly wider than 4:3/16:9 source material (so the PAR stays the same - explained by edDV earlier) or if you know the source has line blanking.

- The ability to scale 4:3/16:9 source straight to 720x480/720x576 as long as sequence_display_extension and display_horizontal_size are present, I'm not so sure on...

That's my personal conclusion, for now.

[jagabo]

If they're picture information they're meant to be seen.

But seen at what aspect ratio? Should the 720x480 image be scaled to 16:9 or to 16.3632:9? No consensus was reached over the use of sequence_display_extension.

I thought the iso13818-2 spec was quite clear:

If sequence_display_extension() is not present then it is intended that the entire reconstructed
frame is intended to be mapped to the entire active region of the display.

[Anonymous344]

That's my personal conclusion, for now.

Apart from Jagabo's last addition, which seems to clarify the matter of the sequence_display_extension, that seems to be more or less where I am.

[intracube]

I thought the iso13818-2 spec was quite clear

Earlier I noted that:
- iso13818-2 covers many applications of digital video (digital satellite,digital cable, digital terrestrial,DVD) as mentioned in the introduction to the pdf document.
- Some of the features listed in the document may not be used in all cases. The wikipedia 'Vob' article states, 'VOB is based on the MPEG program stream format, but with additional limitations'. and 'While all VOB files are MPEG program streams, not all MPEG program streams comply with the definition for a VOB file'
Just because sequence_display_extension is in the iso13818-2 document, doesn't mean it's necessarily in the 'DVD-Video Book' - which I think is the official spec for DVD Video.

My argument raises a question - 'Why *are* there sequence_display_extension headers on some DVDs?' There are two possibilities;
1. They are part of the DVD-Video standard, but they aren't mandatory, so some encoders don't add them.
2. Some encoders used to produce the MPEG-PS streams might be general purpose, and be used in other applications where the sequence_display_extension is used (server playout systems for live digital TV, for example) and they may be incorrectly added when producing video files for DVDs.

From looking at the headers on some DVDs, the sequence_display_extension values made no sense; 540x576 viewable on a 720x576 widescreen DVD? I think that would produce a very distorted image. It looks as though hardware DVD players ignore the values anyway.

For example, on page 50 of the iso13818-2 pdf, there's a chroma_format flag. Table 6-5 has options for 4:2:2 and 4:4:4 colour sampling. I'm pretty sure these aren't allowed for video DVDs - they're listed there as options for other digital systems.

[jagabo]

Just because sequence_display_extension is in the iso13818-2 document, doesn't mean it's necessarily in the 'DVD-Video Book'

I understand what you're saying about DVD not implementing all MPEG 2 features. But I don't see any way of interpreting:

If sequence_display_extension() is not present then it is intended that the entire reconstructed frame is intended to be mapped to the entire active region of the display.

to mean anything other than (without a sequence_display_extension) the full 720x480 MPEG frame represents the 16:9 (or 4:3) picture.

I seem to recall others saying the DVD spec usually refers to the iso13818-2 spec for details, simply noting some exceptions (like frame size limitations). Until I hear reports from someone who has the DVD spec indicating otherwise, I will assume that, in the absence of a sequence_display_extension, the full 720x480 frame represents the 16:9 (or 4:3) picture.

My argument raises a question - 'Why *are* there sequence_display_extension headers on some DVDs?' There are two possibilities;
1. They are part of the DVD-Video standard, but they aren't mandatory, so some encoders don't add them.

I believe this is the case since many commercial DVDs include the field.

From looking at the headers on some DVDs, the sequence_display_extension values made no sense; 540x576 viewable on a 720x576 widescreen DVD? I think that would produce a very distorted image. It looks as though hardware DVD players ignore the values anyway.

This does make sense. On the few DVDs I saw with this the VOB file was flagged as 4:3, not 16:9 (16:9 was obviously the correct final DAR, and the IFO marked it as 16:9). So 540x576 (or 540x480 on NTSC DVDs) represents the 4:3 portion of the frame. Hence the full 720x480 frame is 16:9 (1.333 * 720 / 540 = 1.778).

[ronnylov]

There is a setting called "Pan-and scan" on DVD players that is intended for a 4:3 TV to view 16:9 video zoomed to fill the 4:3 screen. There is some header in the MPEG-2 stream which tells what part of the picture is meant to be viewed on the 4:3 screen in pan-and scan mode. Maybe this is where the 540x576 header comes into account, or maybe somebody have mixed up the headers when they encoded the video?

It looks like this option is supposed to be used for the pan&scan feature.

http://books.google.se/books?id=f6Cuz9Tig-MC&pg=PA213&lpg=PA213&dq=sequence+display+ex...ension&f=false

http://en.wikipedia.org/wiki/Pan_and_scan

Regarding original question of 704x480 16:9 I found this link:
http://www.mpeg.org/MPEG/DVD/Book_B/Video.html

"16:9 (all formats except 352 pixels/line)"

[intracube]

But I don't see any way of interpreting:

If sequence_display_extension() is not present then it is intended that the entire reconstructed frame is intended to be mapped to the entire active region of the display.

to mean anything other than (without a sequence_display_extension) the full 720x480 MPEG frame represents the 16:9 (or 4:3) picture.

I completely agree with you on the definition, and what it means. But, it seems to go against actual examples of DVDs I have:

What's your view on the Goldeneye images I posted?
- The Goldeneye DVD doesn't have sequence_display_extension present
- But if the entire image is displayed at 16:9, the gun barrel is distorted (see image 2)

There is a setting called "Pan-and scan" on DVD players... ...Maybe this is where the 540x576 header comes into account, or maybe somebody have mixed up the headers when they encoded the video?

You're correct; Page 66 section 6.3.12 of iso13818-2 - 'The picture display extension allows the position of the display rectangle whose size is specified in sequence_display_extension() to be moved on a picture-by-picture basis. One application for this is the implementation of pan-scan.'
Not sure, but I thought I read somewhere that DVDs were limited to horizontal pan-scan.

Thanks for all the input, everyone.