General: Aspect Ratio Clarification

[edDV]

edDV: I am kind of reading your explanation bit by bit to understand everything, and I have a question for you regarding PAR


In your example of 352x480 scaling to 640 x 480(the square pixels) I get PAR= 1.8181 You conclude par=1.8182

What I did was:

(640/480) / (352/480) = 1.333333333/0.733333333 = 1.8181 PAR


The earlier explanation of 704x480 scaling to 640 x 480(square pixels) to get Par= .9090 made sense and we BOTH agree on the the same answer:


(640/480) / (704/480) = 1.333333333/1.466666667 = .9090 Therefore PAR = .9090 And we both agree


But why isn't it working above ?

I rounded 1.818181... to 1.8182

[Alex_ander]

I have noticed a certain trend on these boards which is not encouraging: people tend to say "don't bother or just give it up." That is a negative thing. I post questions because I want to learn. And it kind of irritates me when people just say "give it up."

I only meant there is much more simple way of getting the same results (without remembering numbers, actually useless) and didn't suggest you to stop learning. Of course you can follow any other way. Good luck.

[Krelmaneck]

Lanczosresize(720,360)
AddBorders(0,60,0,60)

True, but sometimes it helps that if I can actually know the theory of what calculations to make. I can take my time. But I guess FITcd would be GOOD to double check my work, once I learn the theory.

The theory behind conversions between AR's is quite simple. Let's assume that in your example 576 x 240 represent a film, shot with 2.21:1 AR (it is actually 2.4 from given numbers, but that's just for example) and you want to letterbox it in 16:9 anamorphic encoding. You know that 704x480 represent a 16:9 DVD image. Since 2.21:1 is 'wider' than 16:9, the width 704 will not change and you only want to find out the height of new image in # of samples. From global relationship between the two AR's 16:9/2.21:1=0.804 you get the wanted coefficient. For NTSC the new height will be 480x0.804=386 (576x0.804=~464 for PAL). Round it to a multiple of 16 (if possible, otherwise to a multiple of 8 or 4). Then you get 704x384 and complement it to 704x480 (or 720x480) by addition of borders, then encode with 16:9 setting. As you can see, no need in 'PAR's.
LanczosResize(704,384)
AddBorders(0,48,0,48)

Thanks.

[manono]

Because if rounding is fine, then I can I round the PAR of .9090 to .9091? That is an honnest question of mine. I am not joking.

Well, I'd round 0.90909090909... to 0.9091 myself. Again, though, the difference is less than one ten thousandth. It's not as if using one instead of the other will affect the final resolution or the final aspect error. There will be a much greater aspect error created by using Mod16 or even Mod8 in your final resolutions. You're talking about differences of less than 0.01%, while it's generally said that our eyes can't tell an aspect error of less than 2% when round things begin to look like slight ovals.

[Krelmaneck]

edDV: I am kind of reading your explanation bit by bit to understand everything, and I have a question for you regarding PAR


In your example of 352x480 scaling to 640 x 480(the square pixels) I get PAR= 1.8181 You conclude par=1.8182

What I did was:

(640/480) / (352/480) = 1.333333333/0.733333333 = 1.8181 PAR


The earlier explanation of 704x480 scaling to 640 x 480(square pixels) to get Par= .9090 made sense and we BOTH agree on the the same answer:


(640/480) / (704/480) = 1.333333333/1.466666667 = .9090 Therefore PAR = .9090 And we both agree


But why isn't it working above ?

I rounded 1.818181... to 1.8182

Cool. Thanks.

BTW, are you a video teacher at a university or college?

[Krelmaneck]

Because if rounding is fine, then I can I round the PAR of .9090 to .9091? That is an honnest question of mine. I am not joking.

Well, I'd round 0.90909090909... to 0.9091 myself. Again, though, the difference is less than one ten thousandth. It's not as if using one instead of the other will affect the final resolution or the final aspect error. There will be a much greater aspect error created by using Mod16 or even Mod8 in your final resolutions. You're talking about differences of less than 0.01%, while it's generally said that our eyes can't tell an aspect error of less than 2% when round things begin to look like slight ovals.

Good to know. Now, I have not heard the concept of Mod, is that another ITU kind of strange pixel thing?

[manono]

Mod16 means the resolution's width and height are divisible by 16, Mod8 by 8.

[jagabo]

Because if rounding is fine, then I can I round the PAR of .9090 to .9091? That is an honnest question of mine. I am not joking.

Well, I'd round 0.90909090909... to 0.9091 myself.

Or just multiply by 10 and divide by 11.

And maybe a clarification: mod 16 means in integer multiple of 16 (16*1, 16*2, 16*3...). Or a number that can be divided by 16 leaving no modulus (remainder).

[edDV]

BTW, are you a video teacher at a university or college?

Do they teach video in college? Maybe I should apply.

I was in the industry back when all this was happening 1978-1996 making equipment and participating in the SMPTE standardization process. Since then I've been more of a user of video tech rather than maker.

[Krelmaneck]

I have been thinking really hard about this concept and it has led me to this question: when dealing with a conversion to SVCD, DVD or any medium made to play on your 4:3 tv, is there a time where the pixels are actually square and no resizes are necessary ? It seems like to me that all the time you will have to resize for DVD or even svcd. No medium I know has an image area of 640 x 480 except the tv itself---and even that is a misuse of words because the aspect ratio is supposed to be in inches or cm for the tv. And you will always have a PAR in the mpeg because you are always scaling. Computers are fine with this. So for a computer, yeah, it can have a square pixel video. But it seems like to me you should always resize a dvd to 704 x 480, and then let the DVD do the correct stretching.

Check this out:
640 x 480 the perfect 4:3 square pixel resolution. Now, even though you are dealing with a square pixel in this size, when you author for an mpeg 2 DVD, the image size is 704 x 480, which stretches itself horizontally to 640 x 480 and you get the PAR=.9090. When does par= 1 or 1/1? Never?


Is there any medium that actually has an image data area of 640 x 480, without the need for PAR. The only two I can think of is the TV and the computer screen.


Does maybe a VCR the only medium that does not have to rescale? For example, having a video and playing the data video on the computer and recording it straight to tape via graphics card? People probably will not do that these days. Or does the the tape of an image data area to and stretch a video recorded straight from the computer to video cassette? I am thinking somehow the VCR cannot even record a square pixel if it was from the computer. Now, if I recorded something directly from broadcast to tape, then it can keep the true square pixels maybe?

[manono]

I may not be understanding you correctly, but what about all the millions of Xvid and DivX AVIs out there at 640x480, 512x384 and other 1.33:1 ratios?

[edDV]

All VCR's record by scan lines (vertically sampled) but analog formats have no horizontal pixels, only analog bandwidth. Digital VCR's sample the horizontal lines into pixels. Same goes for transmission systems and digital displays.

Computer hardware and applications evolved with square pixels because there was no real benefit to do otherwise. Video transmission and recording systems optimized sample rate (horizontal pixels) to economic and performance criteria.

Take SD digital cable as an example They could have used 640x480 square pixel frames, but instead chose 8 bit 524x480 as good enough, allowing for less MPeg bit rate (approx 3 Mb/s) for a given signal to noise. This allowed them to carry 20% more digital channels.They concluded that customers valued more channels over slightly better picture quality.

Broadcasters on the other hand needed to produce and transmit video at much higher quality to survive multiple generations of production and transmission recodes. Broadcasters chose 10 bit 720x480 at bit rates above 90 Mb/s (Digital Betacam).

In both cases there was no value to maintaing square pixels.

Manono listed xvid/divx formats at various 4:3 square pixel resolutions. Those formats evolved for display on square pixel computer monitors.

[Krelmaneck]

I may not be understanding you correctly, but what about all the millions of Xvid and DivX AVIs out there at 640x480, 512x384 and other 1.33:1 ratios?

Computer stuff already covered. i meant 640 x 480 videos in any type of MPEg's that do NOT have to be horizontally scaled by your player and thus end up with no PAR or just par=1. Maybe there is a strange mpeg format? True some Divx standalone players such as my Phillips which is Divxx Ultra in which the video can keep it's aspect ratio. But is there any MPEG that plays on a type of MEDIUM/MEDIA that does not have to be rescaled by the player and thus keep it's aspect ratio. (IE. 704 x 480 is horizontally scaled to meet 640 x 480 video, thus a par).



application: DVD has non-square pixels but scales to meet the 640 x 480 video on your tv. Taking it that into affect, I cannot think of any DVD mpeg that will not have to be rescaled. Thus, I was thinking, most mpegs are non-square pixels until the player rescales it.


Get it now?

[jagabo]

MPEG 2 supports 1:1 PAR. So you can encode 640x480 with square pixels if you want. You may have trouble finding anything to play it on TV though. DVD doesn't support it (though I suspect many DVD players will play it properly).

But note that DVD players generally do not perform a digital scaling from 720x480 to 640x480 for analog output. They just adjust the timing of the video DAC so all 720 pixels are output with the correct timing to create a 4:3 image on the TV. This prevents moire artifacts that you would get otherwise.

[Krelmaneck]

MPEG 2 supports 1:1 PAR. So you can encode 640x480 with square pixels if you want. You may have trouble finding anything to play it on TV though. DVD doesn't support it (though I suspect many DVD players will play it properly).

I did not even think that MPEG-2 supported a par of 1:1! That is why I was asking to see if I understood everything because most of the time when you do have a 640 x 480 video source, it would have to be rescaled horizontally to 704 x 480. to fill in the area of the dvd 704 x 480 area.. What medium would support a 1:1 par mpeg besides a computer? i thought it would be nothing, right? And that is why YOU said " So you can encode 640x480 with square pixels if you want. You may have trouble finding anything to play it on TV though" Am I right?

[manono]

It's MPEG-2 that supports 1:1 (and 4:3, 16:9 and 2.21:1). DVD video is a subset of that and supports only 4:3 and 16:9.

You can encode for 1:1 and mux with ImagoMPEG-Muxer or similar, but not with any DVD authoring program, and then test it out in your players. I have no idea which players will play it properly and which won't.

[Krelmaneck]

It's MPEG-2 that supports 1:1 (and 4:3, 16:9 and 2.21:1). DVD video is a subset of that and supports only 4:3 and 16:9.

You can encode for 1:1 and mux with ImagoMPEG-Muxer or similar, but not with any DVD authoring program, and then test it out in your players. I have no idea which players will play it properly and which won't.

To cut a LONG story short just say "Yes. MPEG 2 can support a par of 1:1 but the DVD flag may or may not support it because it is UNNATURAL." And by knowing this information, I can conclude that for the DVD subset or even the VCD/SVCD subsets that a lot of standalone players may not be able to support an mpeg2 DVD or VOB set that are encoded directly at a par of 1:1. and then from that I can conclude to this conclusion: In most cases, when you encode any DVD or SVCD/VCD, it is highly unlikely that it will be able to play on your DVD player with a 1:1 ratio, assuming that the DVD player plays VCD cdr's since there are some standalone that will not. Conclusion: So even though mpeg 2 can support par of 1/1, it is not really "efficient" for DVD or SVCD use. Right, or incorrect?




Edited: 480 x 480 is not even 1:1 on a computer screen. Idea deleted.

[edDV]

MPEG 2 supports 1:1 PAR. So you can encode 640x480 with square pixels if you want. You may have trouble finding anything to play it on TV though. DVD doesn't support it (though I suspect many DVD players will play it properly).

I did not even think that MPEG-2 supported a par of 1:1! That is why I was asking to see if I understood everything because most of the time when you do have a 640 x 480 video source, it would have to be rescaled horizontally to 704 x 480. to fill in the area of the dvd 704 x 480 area.. What medium would support a 1:1 par mpeg besides a computer? i thought it would be nothing, right? And that is why YOU said " So you can encode 640x480 with square pixels if you want. You may have trouble finding anything to play it on TV though" Am I right?

MPeg can have any PAR but players expect standards based resolutions. If you gave an MPeg player square pixels, you would need to go into manual aspect ratio settings.

DVD MPeg2 only supports the previously discussed resolutions and only 4:3 an 16:9 aspect ratios.

So far we have only been discussing SD resolutions. I've been holding off on the HD story until you understand SD.

Looking more generally into computer, video and LCD-TV resolutions you will see different evolutionary tracks for resolutions. This is why "conversion" is so difficult. Adding to the difficulty are all the film aspect ratios.

You have been thinking about a TV set as having square pixels, but many of them don't.

Popular computer square pixel video resolutions
"NTSC" - 1920x1080, 1280x720, 852x480, 640x480, 320x480, 320x240
"PAL" - 1920x1080, 1280x720, 1024x576, 768x576, 384x576, 384x288
VESA Screens - 1920x1080, 1680x1050, 1280x1024, 1024x768, 800x600, 640x480

Common digital video standard resolutions (ITU/EBU and SMPTE)
NTSC/ATSC -
1920x1080*, 1280x720*, 720x480, 704x480, 640x480*, 352x480, 352x240
Note that 640x480 is supported by ATSC but not by DVD or Blu-Ray.

Japanese HiVision (Muse) - 1060x1035, 530x1035

PAL/DVB -
1920x1080*, 1440x1080, 1280x720*, 720x576, 704x576, 352x576, 352x288

* square pixel

720x480 and 720x576 are supported for DVD and Blu-Ray but not for ATSC/DVB.

Common LCD/Plasma TV screen resolutions
1920x1080, 1366x768, 1280x768, 1024x768, 960x768, 800x600, 640x480

[Krelmaneck]

deleted by user. rethinking post

[Krelmaneck]

I know where I screwed up. I was constantly thinking of having a 704 x 480 dvd image area scaling for a 640 x 480 video, and now that number did not get out of my head. I was thinking then most pixels have PAR from doing those conversions and that is why I asked the question about feeding a direct 640 x 480 video into a 1:1 scenario hoping most DVD players would reject it thus proving DVD will not play an mepg 2 for DVD if it is not re sized to one of its standards for DVD sizes, thus proving that a PAR of 1:1 is ok for an mpeg but NOT OK for a mpeg 2 DVD. .


And then, I thought the total opposite for one second because I thought 1.3333=square pixels. And the reason why I was thinking of square pixels was because I trained my brain to think a DVD video had to h scale from 704 x 480 to 640 x 480 but that has NEVER been the case. It was just that darn example I brought up of MY video being 640 x 480 square pixels. That is where I screwed up. And I was thinking of edDV's first reply to my thread of sampling 480 lines at a 4:3 aspect ratio. And from there on I thought it was gospel that 640 x 480 must be NTSC's only sampling matrix.




Ok. So maybe I need to think of a new example so I will not think DVD=square pixels for NTSC 4:3. So let us get rid of 640 x 480 out of my head.


Let us think of 480 x 480 but NOT for SVCD but scaling from this to DVD. Maybe if I think this way, I will stop thinking 1.333=square pixels, and begin to think of PAR one more time and get back on track.



I know I confused the other people responding to my other post because of my 640 x 480 example, and that is why someone said that DVD's do not do the resizing. I got answer like this because I forgot to tell the responder I had a video of 640 x 480, and that is why the responder responded the way he did.


Let us start anew now with some new questions:


1 will start by saying: thanks for posting the standards for NTSC.




How about a new example of 480 x 480 resizing to DVD 4:3 NTSC.




Ok. that is all I am going to post for tonight. My brain is hurting.

[edDV]

Here is the part where I am getting confused now. Let us try to just stick to an NTSC signal. This is what I am not understanding. I keep on thinking that a DVD video must be scaled back to the square pixels because that is what the screen likes--square pixels. Otherwise, if TV did not care about using square pixels then we could just let the DVD set the standards for 720 x 480 without a need for a player to auto-h scale to the square pixels.

The only TV type with a 640x480 pixel frame buffer is the 4:3 type I referenced above.

The general case for a DVD player is analog NTSC out (composite or S-Video) or analog components (YPbPr) out. If the DVD player TV setting is for a 4:3 TV and the DVD disc is flagged 4:3 (default), then 720x480, 704x480 or 352x480 digital video will be digital to analog converted in such a way that each digital line horizontally fills the analog scan line giving a 4:3 analog signal. This analog NTSC signal will be displayed as 4:3 on old analog or new digital TV sets.

If the DVD disc is flagged wide, the processing gets more complicated but the result is a letterbox or cropped Pan/Scan image in a 4:3 analog NTSC frame.

If the DVD player is connected HDMI to a digital TV, the digital TV will process the 720x480, 704x480 or 352x480 input to display 4:3.

[edDV]

So maybe I need to ask myself this question: Is there a dvd standards image area that when rescaled to 4;3 aspect ratio NTSC where they are actually are not square pixels but still does maintain the 4:3 aspect ratio? Can someone answer that question? Maybe I have the improper thinking 4:3 areas must be square pixels for NTSC. So let's take the 352 x 576 standards for DVD. So let's say you have a 352 x 240 video and you wanted to H scale for DVD, would the video have square pixels or would it be non-square pixels?

Above we were talking about how a standard DVD disc plays. Now I think you are asking about encoding video for DVD so that it plays as 4:3 on a DVD player?

If the 352x240 video is 4:3 aspect ratio, it not square pixel.

4:3 square pixel versions would be 320x240 or 352x264.

If the 352x240 video is 4:3, it is DVD ready as is. Just encode it. 352x240 is an approved DVD resolution. The DVD player will output it as analog 4:3.

352x240 is the same thing as VCD. It represents quarter resolution 704x480.

[edDV]

480x480 square pixel would be a square frame, not 4:3.

If 480x480 represented a 4:3 image, it would appear horizontally squeezed if viewed on a square pixel display. This is exaxtly how DirectTV and Dish first broadcast SD Mpeg2 material from the satellite to the home. The tuner converted this signal to analog 4:3 NTSC.

To convert 4:3 480x480 to DVD, you would resample or stretch each horizontal line from 480 to 704. The result would be 704x480, a standard DVD resolution.

[jagabo]

Just FYI: The MPEG 2 spec uses a numeric flag to indicate aspect ratio:

flag value = aspect ratio
1 = 1:1 PAR (square pixel)
2 = 4:3 DAR
3 = 16:9 DAR
4 = 2.21:1 DAR

A 640x480 4:3 DAR MPEG frame could be flagged as PAR=1:1 (1) or DAR=4:3 (2) -- the result is the same, 4:3 DAR.

Analog video has no pixels. It is a continuous waveform.

[Krelmaneck]

Ok. I am kind of off the beaten track here.


Let us talk about MY DISPLAY.


edDV

My TV is an old emerson TC1972d 19" television that is 4:3. I have NO clue what the resolution actually is. All I know is that is 4:3. My DVD player is a Phillips that can HD upscale with HDMI output, BUT is outputted by analog (rca cables) to the line in of my GO VCR, and the VCR is connected to my Emerson TV via coax (75 ohm). I am thinking my TV must be analog since it was like almost 20 years old.


So now, let us just think about MY TV and the the way the the video signal would be displayed on my TV.


So, let us say for example, I encoded a 480 x 480 video and resized to 704 x 480 for DVD, which you told me to do. Now, thinking about how my TV is set, what device would be doing the upscaling? Bare in mind, we are talking about how the dvd player plays it. (bad sentence LOL) I know I do not have and HDMI tv--I just said that to illustrate I have a DVD player that is less than 6 years old. Anyhow, since my tv is analog I am thinking would I be correct in thinking that the Phillips DVD player would take the 704 x 480 video and upscale it to the 640 x 480? We are talking about the actual upscaling that is done either by the TV or the DVD player and not the process of encoding but how the DVD or TV set would upscale this assuming that the DVD disc identifies itself as 4:3. Like in my case would the DVD or TV player upscale the video from the not-square pixels to the square pixels? and for ALL SVCD and VCD's, would the tv player H scale/upscale to square pixels only on MY DISPLAY?

.

[jagabo]

So, let us say for example, I encoded a 480 x 480 video and resized to 704 x 480 for DVD, which you told me to do. Now, thinking about how my TV is set, what device would be doing the upscaling?

You are upscaling from 480x480 to 720x480 in software before making the DVD.

Bare in mind, we are talking about how the dvd player plays it. (bad sentence LOL) I know I do not have and HDMI tv--I just said that to illustrate I have a DVD player that is less than 6 years old. Anyhow, since my tv is analog I am thinking would I be correct in thinking that the Phillips DVD player would take the 704 x 480 video and upscale it to the 640 x 480?

First of all, that would be downscaling, not upscaling. And no, the digital image is not downscaled from 720x480 to 640x480 anywhere. The DVD player decompresses the video to a 720x480 frame buffer. It then uses its video DAC (digital to analog converter) to output one pixel every ~74 nanoseconds creating a continuous analog waveform that the TV scans from left to right across the CRT screen. The amplitude of the waveform corresponds to the brightness of the electron beam as it is scanned across the face of the TV.

in my case would the DVD or TV player upscale the video from the not-square pixels to the square pixels? and for ALL SVCD and VCD's, would the tv player H scale/upscale to square pixels only on MY DISPLAY?

There is no upscaling or downscaling going on. There is just different sampling rates of the digital data as it is sent to the DAC. SVCD will sample the 480 pixels at a rate of about 112 nanoseconds per pixel, VCD at about 148 nanoseconds per pixel. (It's possible that some DVD players upscale the 480x480 SVGA, or 352x240 VGA digital frame to 720x480 (or 704x480) before sending the pixels to the DAC at standard DVD rates.)

You have to stop thinking of pixels when dealing with analog TV. There are no pixels there. Pixels only exist in digital video.

[edDV]

Your Emerson TV is an analog NTSC TV and knows nothing about pixels. It receives the standard analog NTSC signal from the DVD player. All the action takes place in the DVD player itself. The analog NTSC signal is similar to that produced by the now defunct analog TV stations. If you don't understand that part, see here.

http://en.wikipedia.org/wiki/NTSC
http://www.ntsc-tv.com/

The DVD player converts digital data on the DVD disc into an analog NTSC signal.

[Krelmaneck]

If an SVCD was encoded at 480 x 480 as MPEG 2 with a dar 4:3, since my current analog tv does not deal really with pixels, why would the beam(?) or waveform kind of incorrectly show the picture wrong on my TV with some of the picture cut off the sides (the left and right)? Is possible because the DVD player tried to stretch out the 480 x 480 video to 720 x 480 before sending it to the DAC?

[jagabo]

If an SVCD was encoded at 480 x 480 as MPEG 2 with a dar 4:3, since my current analog tv does not deal really with pixels, why would the beam(?) or waveform kind of incorrectly show the picture wrong on my TV with some of the picture cut off the sides (the left and right)? Is possible because the DVD player tried to stretch out the 480 x 480 video to 720 x 480 before sending it to the DAC?

It's possible but it's more likely the DVD player adjusting the DAC's timing to create the correct analog waveform from the 480x480 frame buffer. Resizing the frame buffer from 480x480 to 720x480 would be a waste of time and create digital scaling artifacts.

[Krelmaneck]

NTSC Analog and DVD ADR:


Analog TV:

I found an article on the internet that described an analog TV as having a vertical and horizontal resolution. The Vertical resolution would be the number of horizontal lines scanned from left to right and then counting down from top to bottom. Thus, this would be the height of the display which never changes--it is fixed. That makes sense. And then it described the horizontal resolution as counting the number of vertical lines, thus getting the displays Width, which changes dependent on the video's signal. The thing that confused me in the article is that it said DVD had a horizontal resolution as having 480 lines, VHS as having 240 lines and broadcast having 330 lines. My question is: if the Vertical resolution is fixed at 480(or 486) and the Horizontal resolution of DVD is 480, how does that 480= 720 x 480 or even 640 x 480. The vertical resolution is fixed in an analog signal, so need to explain that. what I am not getting is the Horizontal Resolution. What would make sense to me would be Width 480(dvd horizontal resolution) x 480(fixed analog Vertical resolution)= 480 x 480. what am I missing?


My other question is: The article also mentioned in order to caluclate resolution, you do so by calculating it by the number of pixels per scanline. How do you know how many pixels are in a scanline? The only thing I can think about is number of Horizontal Resolution factor (the width that changes based on video signal, and that is the number of pixels per scanline, I probably wrong though. What is it?