Export HD to NTSC Widescreen -- not truly 16:9

[Sengstack]

If you export HD footage, 1080x1920 for example, to NTSC Widescreen, you end up with 720x480 with rectangular pixels that have a 1.2121 pixel aspect ratio. That means, the display is effectively 873x480 pixels, which is a a screen aspect ratio of about 16.4:9 instead of 16:9. What you end up with is the original video image squashed between thin, vertical black bars on the right and left sides (not the thick black bars you get when you display a 4:3 video in a 16:9 display).

If you open the video in an NTSC Widescreen Sequence, you'll see those black bars. The video image does not fill the screen.

This is easily fixed: change the Scale Width value to 102.3 (uncheck Uniform Scale). That puts the image portion of the clip into the proper aspect ratio (and shoves the black bars off the left and right sides of the screen). But I'd rather change the scale of all the clips in a project.

Since NTSC has to be 720x480 pixels, if you want to display something at a 16:9 ratio, the pixel aspect ratio should be 1.185 instead of 1.2121.

Perhaps an Adobe engineer can explain why exporting HD to NTSC Widescreen creates a video with black vertical bars and how a 720x480 (with rectangular pixels with a 1.2121 aspect ratio) clip can be considered as having a 16:9 aspect ratio.

Thanks,

Jeff Sengstack

[Sengstack]

Ann wrote: Actually the black bars is empty space. The background shows as black.

==========

Not so in my experience. I just exported an AVCCAM (AVCHD) 1920x1080 30p clip to NTSC DV Widescreen and the bars along the left and right sides are black, not transparent.

Jeff

[Wil Renczes]

Jeff, I already answered your question above (not sure if you skipped over my post).  You're getting tripped up on the notion that the black bars equates to the area outside the production aperture of dv footage, and that's not the case.  This problem has nothing to do with production versus clean aperture of dv.

(Tangentially, since you were asking on what you're getting during onscreen dv playback , PPro doesn't expand out production aperture to clean on playback - that would be effectively be a horizontal stretch distortion.  You're getting full 720 width in the program monitor.  If you're wondering why you're not seeing black bars on your DV  footage, it's probably because there isn't any.  Keep in mind that the 'production' aperture might not be visible on a dv camera - you'd have to be capturing from an analog signal to see production aperture.  If you're shooting on a DV device, it's likely to fill the entire width of the available image.)

To repeat, the black bars are the difference between your HD aspect ratio (usually 1.5 or 1.33) and the NTSC 16:9 aspect (1.2).  To maintain the 1:1 relationship of the height & width of your HD material, letterboxing is being applied.

[Sengstack]

Hi Will, et al,

I'm throwing in the towel. It's making less sense now than after the first couple responses.

Back then, as I understood it, the issue was that clean aperture was the visible portion of the larger production aperture. And the area outside the clean aperture but inside the production aperture was not displayed in P Pro. Now Will says "P Pro doesn't expand out production aperture to clean on playback." How could P Pro expand out production aperture to clean aperture, when clean aperture lives within the production aperture? At least, until now, I thought the clean aperture was within the production aperture. Now I don't know.

Will also wrote in his earlier post about different PARs. He said the "OP question" (I don't know what OP question means) "is why do I get letterboxing?" I don't see what difference, different PARs make. Working with different PARs happens all the time. How about working with 1280x1080 with a 1.5 PAR? If that is output to 1920x1080 with a 1.0 PAR, do the PAR differences create an problem? And PARs for NTSC have been shifting around as engineers figured out that 10/11 is better than .9 and so on.

Anyway...I had hoped to be able to figure this out. But now it's clear that short of sitting down with an Adobe Media Encoding software engineer with P Pro and AME open in front of us and a pocket calculator standing by, I don't think I'll be able to make heads or tails of this.

Jeff

[able123]

hehe...When you do that...sit down and go through this with SMPTE and NASA and MIT and ADOBE engineers and programmers ----SHOOT VIDEO....document it and edit with graphics and screenshots and math forumulas etc....

Then you can bet I will be on line to BUY IT   !

[Sengstack]

Will do. Thanks for the chortle.

[FastFeet · Accepted solution]

Hi Jeff,

I just run accross your post. It is old, but I recently was running in the same problems as you. It was hard to get things straight for me, but finally I found a German paper that explained everything just to the point.

16:9 = 1,777...

HD uses square pixels (PAR=1.0) and if you check the two most common HD video sizes and take your calculator 1920:1080=1,777... and 1280:720=1,777.... So both of them are 16:9 formats. (Exactly we would calculate 1920 (pixel) x 1.0 (PAR) : 1280 (pixel) x 1.0 (PAR))

Now in SD we have an analog NTSC/PAL an a DV NTSC/PAL. To make it easy first, lets look at DV. The default resolution for DV NTSC is defined as 720×480, but DV doesn't us square pixel. It uses rectangular pixel. The PAR for analog(!) 16:9(!) NTSC is about 1,2154. So to convert DV pixel into the square pixel world you need to calculate 720 (pixel) x 1,2154 (PAR) = 875 square pixel.

Now we are ready for the surprise: 875:480 = 1,822916...!!! DV NTSC (as PAL) are just not exact 16:9!

This means you can't transform an HD frame into an SD DV frame without cropping or padding with black bars.

Old Premiere and AE actually stretched the frame a little to make the transform fit. In AME you really need to choose one of three options: you can have the HD frame cropped, the SD frame padded with black or the SD frame stretched, which would distort the picture a little bit, but not much. If you don't want to have the picuedistorted or have black bars you need to crop off 28 (27) pixels from your HD surce footage. This is why often 14 pixels are each cropped of the top and the bottom. You can do this in AME using the crop tool and set the output "scale to fit".

So far this explains your problem and shows you the possible solutions. To understand why DV NTSC and PAL are not 16:9 you have to dig a little bit deeper into the analog world. This would take bit longer. For a rough and simple explanation consider that analog TV (NTSC/PAL) has no pixels. It's analog. So ITU-R BT.601 defines how to get analog video into the digital world. The defined sampling frequency results in a NTSC frame of 711×486. Considering the PAR of 1,2154 you get 864:486=1,777.... So analog (anamorphic 16:9) PAL is really 16:9. Analog NTSC video convertd to DV is padded with black bars to fill to 720.

This all concerns analog to digital SD-TV. The black bars were not shown on CRT TVs due to the overscan area. Nowadays LCD TVs and computer monitors can show the full 720 pixel, but still use overscan. Also video that is recorded digitally is normally recorded with the full 720 pixel. So we have video that is not in 16:9. Therefore you need to handle the transforming either with bars, cropping or distorting.

Hope this helped you Jeff.

Marcus

[Sengstack]

Hi Marcus,

Thanks for the thorough and informative explanation.

Jeff

[HarleyTDavis]

I've seen the phenomena in broadcast.  The math works too.  Here's a little solution I came up with... ...PADD your SOURCE...  Let me explain.  If we have the black bars coming in anyway, as a matter of course, why not simply add that to the source and downscale it a bit before dropping it down to SD output?  This puts a Padding on it in the source area, and cropping afterward doesn't lose footage at the edge.  You may have actually solved a problem I'm dealing with right now with your explanations.  I may go ahead and do this tomorrow when I have time.  You've certainly given me much to think about.

My problem is this:

I am multi-cam editing a 1080 but I want to do so in a 480\486 format so I can output to dvd in a full 16;9(okay, not quite 16;9) format.  I can live with some overscan that gets scaled up when played on a blu-ray player or into a newer HDTV.  I'd like it to be able to play on a 4;3 screen as a panoramic format.  Like I said, I can live with the scaling and padding, no problem.  I prefer it actually.

My solution for you:

set up your SD sequence, note the size by pixel...  Remember, approximately 2.3% of your image will be padding.  Why not downscale that first in an HD sequence (you know, drop 2.3% in HD instead of blowing up in SD)?  When you crop your image later, you'll lose a little bit of the quality in the image right?  Sure... ...But you introduced noise to begin with in downscale and it should be sharper, and when you crop to upscale it back, you should put even more noise into it without asking... ...What to do?!  Soo much noise...  How about adding a 0.5 or 1% gaussian blur in the output step to wash away the excess noise?  You could run a noise filter before, but that won't get you as accurate a result in the output.  Run the gaussian blur in output at less than 1% (this equates to a sharpening filter with a denoiser in photoshop; basically, blurring the image at less than 1% means that it will blur details that vary extremely little, while leaving larger variances intact.  The smaller the decimal value, the smaller the change in detail it will seek.  At 1, it will smooth out pixels where the change in color is a small, and at greater values than 1, it will seek out more of those greater changes;  Add a decimal to the whole number and it will do a color\luma pass like a standard decimal value, but it will find more of those small changes in the areas it has just smoothed over).  Set your Gauss canon, fire...  You should get a decent quality, better than normal SD, but not quite HD.  The motion might look more natural too.

Screw it... ...I'm gonna give this a try when I finish my project.  I'll just make a separate set of sequences if you don't mind... ...It would make a great Special Full Frame Edition, but really wouldn't have a great purpose beyond that.

Just one thing... ...Does anybody know how to get a decent copy of my Markers from one sequence to another, without simply Duping the sequence?  Me neither.  Should work nicely for this one.    All I really have to do is dupe the sequence, replace one video with another, and remake all my cuts.  I'll have to mark each one.

[able123]

I guess you're exporting frames and opening in photoshop to measure exactly how many pixels are left and right of frame? which is square pixels..psd...

I think the bottom line is that what you see is the result of un-exact measurements...relating to par, dar etc.

And a lot of different cameras, who basically translated the japaneese company "standards" of HD ( blue ray etc) according to their not exact language translation...meaning there's a lot of leeway about application of the standard...You would think that translation of science would be more exact but it isn't.

If I go from metric to feet and inches, or nautical miles to statute miles, there are discrepancies ...as the math just isnt perfect.  There is some slop in the math, no matter what you do, and the accuracy of math is sometimes in the translation of the different "systems".

and this is made more obvious as you translate more than 2 things....like maybe now you are dealing with 3 things..

Its a fascinating subject though...thanks for sticking it out and trying to get a definitive answer !

[Sengstack]

Perhaps there's one other part of the equation that has not been factored into this conversation:

Do P Pro, other NLEs, and software video players treat captured DV AVI files differently than transcoded DV AVI files?

What I'm saying is:

1) When P Pro or other NLEs capture DV they put the unchanged DV in an AVI wrapper (presumably this applies to MOV wrapped files as well). So, when P Pro, other NLEs, and software video players "see" that captured DV AVI file, they "know" it started it's life as DV captured from a tape, so they do not display the right and left 8-pixel wide columns and use an appropriate pixel aspect ratio value to shrink or expand the width of the remaining 704 columns to display that visible area properly as SD or Widescreen.

2) On the other hand, how do P Pro or other NLEs display AVI (and presumably MOV) files that were transcoded rather than captured? I'm reasonably certain that transcoded DV AVI files have image information in all 720 columns. That is, I believe they do not have two black, 8-pixel wide columns on each side that are not supposed to appear in the display. If that's the case, do P Pro, other NLEs, and software video players treat those transcoded DV AVI files differently? That is, do they display all 720 columns?

If it is correct to say that there are two types of DV AVI files and P Pro and other software products display them differently, then there are two follow-on comments/questions:

1) It appears that when P Pro  (Adobe Media Encoder) exports HD to NTSC DV Widescreen, it "assumes" P Pro will treat that DV AVI file as if it were a captured DV video clip. So it adds 8-pixel wide black columns on each side of the visible image that, presumably, P Pro will not display (Note: P Pro does not add black-bars to NTSC DV SD files, only to Widescreen).

So, perhaps the HD to NTSC DV Widescreen Media Encoder transcoding algorithm is not working properly.

2) If it is correct to say that there are two types of DV AVI files, then as an update to my previous note...I'm thinking P Pro should interpolate 1920 columns down to 720 (not 704 as I originally wrote before coming to the conclusion that there might be two flavors of AVI files).

From my perspective, it would be good if an Adobe P Pro software engineer, who worked on the Media Encoder and who knows how P Pro displays captured DV AVI files and transcoded DV AVI files, would weigh-in on this.

Jeff

[Jim_Simon]

digital video (DV) was originally destined for analog broadcasting to analog TV sets; to compensate for that, the whole production vs. clean aperture thing got swirled into the mix.

It may have gotten swirled in previous to digital broadcast and displays, but Adobe added it after the fact.  Now that the entire pipeline is digital, that seems as good a time as any to drop some of the old analog ways and develop new standards that are more appropriate to the digital pipeline.

I'm not sure Adobe did the right thing here by conforming DV to old, analog standards.  To wit:

"Although the ATSC A/53 standard limits MPEG-2 transmission to these 18 formats (and their 1000/1001-rate slowed-down versions), the U.S. Federal Communications Commission declined to mandate that television stations obey this part of the ATSC's standard. In theory, television stations in the U.S. are free to choose any resolution, aspect ratio, and frame/field rate, within the limits of Main Profile @ High Level. Many stations do go outside the bounds of the ATSC specification by using other resolutions – for example, 720 × 480."

[able123]

-----------------

you have to do your frame aspect ratio calculations based on the image  area (clean aperture) not the production aperture.

---------------------

page 4...this link

http://www.panavision.com/publish/2007/12/10/GenesisFAQs20071207.pdf

-----------

My take on this is that CCD's that are full size for 35mm film ( beginning of aspect ratios for a lot of stuff ) there is less fudging re: transforming a smaller chip size to the correct exact image size, that the image size derrived from the real camera recording may not be exactly the "production" aspect ratio sizes...or something like that... It is confusing to me also, as I have no video cameras, digital or otherwise, and don't shoot anything but still film.

I think ( not sure but think I read ) that the red camera actually has a CCD "larger" than the genesis, so it actually is capable of recording higher resolutions than typical 4.4.4 production dimensions....but how that works into "clean" image area and aspect ratio is confusing to me also...

To further confuse me, this article has info on exact number of lines broadcast for pal and ntsc...

note THIS in the article ===------------------------------------------------------

NTSC Video

• 525 scan lines per frame,  30      frames per second (or be exact, 29.97 fps, 33.37 msec/frame)
• Interlaced, each frame is      divided into 2 fields, 262.5 lines/field
• 20 lines reserved for control      information at the beginning of each field
  • So a maximum of 485       lines of visible data

---------------------------------------------------------------------------------------

Basics of Video

• Analog video is represented      as a continuous (time varying) signal.
• Digital video is represented      as a sequence of digital images.

    Types of Color Video Signals

• Component video --      each primary is sent as a separate video signal.
  • The primaries can       either be RGB or a luminance-chrominance transformation of them (e.g.,       YIQ, YUV).
  • Best color       reproduction
  • Requires more       bandwidth and good synchronization of the three components
• Composite video --      color (chrominance) and luminance signals are mixed into a single carrier      wave. Some interference between the two signals is inevitable.
• S-Video (Separated      video, e.g., in S-VHS) -- a compromise between component analog video and      the composite video. It uses two lines, one for luminance and another for      composite chrominance signal.

    Analog Video

The following figures are from A.M. Tekalp, "Digital video processing", Prentice Hall PTR, 1995, NTSC.

NTSC Video

• 525 scan lines per frame, 30      frames per second (or be exact, 29.97 fps, 33.37 msec/frame)
• Interlaced, each frame is      divided into 2 fields, 262.5 lines/field
• 20 lines reserved for control      information at the beginning of each field
  • So a maximum of 485       lines of visible data
  • Laserdisc and S-VHS       have actual resolution of ~420 lines
  • Ordinary TV -- ~320       lines
• Each line takes 63.5      microseconds to scan. Horizontal retrace takes 10 microseconds (with 5      microseconds horizontal synch pulse embedded), so the active line time is      53.5 microseconds.

Digital Video Rasters

• Color representation:
  • NTSC uses YIQ color       model.
  • composite = Y + I cos(Fsc       t) + Q sin(Fsc t), where Fsc is the frequency of color subcarrier

PAL Video

• 625 scan lines per frame, 25      frames per second (40 msec/frame)
• Interlaced, each frame is      divided into 2 fields, 312.5 lines/field
• Uses YUV color model

    Digital Video

• Advantages:
  • Direct random access       --> good for nonlinear video editing
  • No problem for       repeated recording
  • No need for blanking       and sync pulse
• Almost all digital video uses      component video

Chroma Subsampling

• How to decimate for      chrominance?

• 4:4:4 --> No chroma      subsampling, each pixel has Y, Cr and Cb values.

4:2:2 --> Horizontally subsample Cr, Cb signals by a factor of 2.

4:1:1 --> Horizontally subsampled by a factor of 4.

4:2:0 --> Subsampled in both the horizontal and vertical dimensions by a factor of 2. Theoretically, the chroma pixel is positioned between the rows and columns as shown in the figure.

• 4:1:1 and 4:2:0 are mostly      used in JPEG and MPEG (see Chapter 4).

CCIR Standards for Digital Video

(CCIR -- Consultative Committee for International Radio)

                       CCIR 601       CCIR 601         CIF         QCIF

                        525/60         625/50   

                         NTSC         PAL/SECAM       

--------------------  -----------    -----------   -----------  -----------

Luminance resolution   720 x 485      720 x 576     352 x 288    176 x 144

Chrominance resolut.   360 x 485      360 x 576     176 x 144     88 x 72

Color Subsampling        4:2:2          4:2:2         4:2:0        4:2:0

Fields/sec                60             50            30           30

Interlacing               Yes            Yes           No           No 

• CCIR 601 uses interlaced      scan, so each field only has half as much vertical resolution (e.g., 243      lines in NTSC). The CCIR 601 (NTSC) data rate is ~165 Mbps.
• CIF (Common Intermediate      Format) -- an acceptable temporary standard
  • Approximately the VHS       quality
  • Uses progressive       (non-interlaced) scan
  • Uses NTSC frame rate,       and half the active lines of PAL signals --> To play on existing TVs,       PAL systems need to do frame rate conversion, and NTSC systems need to do       line-number conversion.
• QCIF -- Quarter-CIF

ATSC Digital Television Standard

(ATSC -- Advanced Television Systems Committee) The ATSC Digital Television Standard was recommended to be adopted as the Advanced TV broadcasting standard by the FCC Advisory Committee on Advanced Television Service on November 28, 1995. It covers the standard for HDTV (High Definition TV).

Video Format

The video scanning formats supported by the ATSC Digital Television Standard are shown in the following table.

Vertical Lines	Horizontal   Pixels	Aspect Ratio	Picture Rate
1080	1920	16:9	60I 30P 24P
720	1280	16:9	60P 30P 24P
480	704	16:9 &   4:3	60I 60P 30P   24P
480	640	4:3	60I 60P 30P   24P

• The aspect ratio for HDTV is      16:9 as opposed to 4:3 in NTSC, PAL, and SECAM. (A 33% increase in      horizontal dimension.)
• In the picture rate column,      the "I" means interlaced scan, and the "P" means      progressive (non-interlaced) scan.
• Both NTSC rates and integer      rates are supported (i.e., 60.00, 59.94, 30.00, 29.97, 24.00, and 23.98).
• At 1920 x 1080, 60I (which      CBS and NBC have selected), there will be 1920 x 1080 x 30 = 62.2 millions      pixels per second. Considering 4:2:2 chroma subsampling, each pixel needs      16 bits to represent, the bit rate is 62.2 x 16 = 995 Mb/sec.

Homepage of the Advanced Television Systems Committee (ATSC)

[Wil Renczes]

Long thread.  If I can distill it down correctly, the OP question is essentially asking 'why do I get letter boxing?'

Essentially, the problem comes down to the fact that your source doesn't match the destination PAR.  Currently, if the source & destination PARs don't match, you get letter boxing.  Why? From a technical standpoint of 'do no harm', it's bad practice to drop pixels, and the render pipeline is trying to preserve the aspect ratio of your footage.  What you're effectively asking for is a option in AME to 'fit to fill (distort)'.

Currently, as a workaround, if you plan to export to a DV preset, if you nest your HD sequence into a DV sequence, then export the nested version, now your source sequence will match the destination, and that will circumvent the need to letterbox.

Cheers

[able123]

Wil,  which is basically putting the HD into DV and scaling it down to fit the 16:9 DV..correct ?

[Wil Renczes]

robodog2 wrote:

Wil,  which is basically putting the HD into DV and scaling it down to fit the 16:9 DV..correct ?

Yep, exactly. Minor nuisance, I know, but at least you know exactly what you're getting for output, since you now control the panning.