Anything to gain by using higher resolution than the DVD?

[scharfis_brain]

what is 4mv ?!?

[bleo]

@scharfis_brain & SeeMoreDigital
Sure all software MPEG2 players use 'generic' PAR, and I don't blame them because it makes the resizing much easier, and results in a convenient 1024 display. :)

scharfis_brain: Are you sure, that DV and ITU PAR aren't the same? (Any literature for that?) I would say, that they are the same.

I just made up the name DV; it doesn't refer to any standard :p. But if you look in xvid and WMV, you will see that the preset PARs for anamorphic encoding use this method.

SeeMoreDigital: Besides... it makes no logical sense to resize an Mpeg2/DVD's 720 image (with edge to edge active pixels) down to 704 prior to encoding

Actually, to follow the DV method, you would CROP down to 704 (regardless of whether the edges have active pixels or not) then resize up to (exactly) 1024 pixels. Or if you wanted to keep the full 720 pixels, you would have to resize up to 1047.27 rounded up to 1048. So once again, things get complicated :p

Anyway, my recommendation is to encode anamorphically using PAR flags: (which vaguely brings us back to the original topic of this thread... :p)
1. Keep the original resolution of the DVD and do not resize.
2. You may crop off any black bars or noisy edges if you wish.
3. Set the PAR in xvid or WMV to whichever method you prefer. In fact, since it is just a flag it should be easy to change even after you encode.
4. That's it! No messy calculations!

btw, there is a THX test on the Star Wars DVD that has circles to test anamorphic encoding...

[darkavatar1470]

After some hard searching, I found a page that explains almost everything, and is the only one that actually tell you how to calculate 702/704 active pixels from the spec.--which I blew my head off trying...

http://www.uwasa.fi/~f76998/video/conversion/#3.2.1

and after some circle drawing,
i found out that my current project (NTSC 16:9) should be cropped to 704 or use a PAR of 40:33...
but i'd like to keep the extra horz. and encode full screen at 1024x564

[SeeMoreDigital]

Originally posted by bleo
Actually, to follow the DV method, you would CROP down to 704 (regardless of whether the edges have active pixels or not) then resize up to (exactly) 1024 pixels. Or if you wanted to keep the full 720 pixels, you would have to resize up to 1047.27 rounded up to 1048. So once again, things get complicated :pActually the correct method would be to crop the image area down to 704 and put it within a 720x576 frame. And introduce DAR signalling. Which is how it "should" be done on DVD and DVB... but more often than not you'll only see 704 laid over 720 when viewing 4:3 transfers and DVB broadcasts. You can see this method being used everyday when viewing Sky News.

Originally posted by bleo
Anyway, my recommendation is to encode anamorphically using PAR flags: (which vaguely brings us back to the original topic of this thread... :p)
1. Keep the original resolution of the DVD and do not resize.
2. You may crop off any black bars or noisy edges if you wish.
3. Set the PAR in xvid or WMV to whichever method you prefer. In fact, since it is just a flag it should be easy to change even after you encode.
4. That's it! No messy calculations!Try using the DAR flags instead of the PAR ones

Originally posted by bleo
btw, there is a THX test on the Star Wars DVD that has circles to test anamorphic encoding... This is very true.

Unfortunately, because the backgrounds of both the 4:3 and 16:9 "circle" samples are black, it's impossible to tell if they've been encoded as "704 laid over 720".


Cheers

[bleo]

@SeeMoreDigital
I think you've missed my point. I'm not concerned about whether or not the edge of the image has black or active pixels, or whether this indicates ITU compliance. This would only be important for display on TVs.

I only use the ITU word to denote its method for defining PAR. Its relevance is that when viewing an encode on a computer, you must use the PAR to convert to square pixels. This is done by either:
1. applying the PAR to the source by resizing, then encoding in square pixels; or
2. encoding the source anamorpically using PAR signalling, then resizing on playback.

Each of the three PAR methods I have defined will result in different images. The ITU method will look fatter than the DV method, which looks fatter than the generic method.

For example, a 720x576 PAL 16:9 DVD resized using the generic method will be 1024x576, DV method: 1048x576 and ITU: 1050x576 (or proportional to these, without regards to cropping). You can try this with the circles in the THX test to determine which method is correct for those DVDs.

SeeMoreDigital: Try using the DAR flags instead of the PAR ones

As far as I know, WMV doesn't use DAR flags, and in xvid, even though you may set DAR, I think it converts and stores it as PAR.

[Sagittaire]

Metric test for 1280*720 screen ...

Source: Video HD-WMV 720p from M$ Amazon_720.wmv

Video 1: 1280*720 encoding with q2 MPEG4 ASP. Bitrate ~6000 Kbps.

clip = DirectShowSource(Source)
clip = Lanczos4Resize( clip, 720, 576)
clip = Lanczos4Resize( clip, 1280, 720)
Return clip

Video 2: 720*576 encoding with q2 MPEG4 ASP. Bitrate ~4000 Kbps.

clip = DirectShowSource(Source)
clip = Lanczos4Resize( clip, 720, 576)
Return clip

Video 3: 1280*720 encoding with q3 MPEG4 ASP. Bitrate ~4000 Kbps.

clip = DirectShowSource(Source)
clip = Lanczos4Resize( clip, 720, 576)
clip = Lanczos4Resize( clip, 1280, 720)
Return clip


Source DVD 576p vs q2 720p encoding

Playing simulation of 576p DVD on 720p screen : SSIM = 90.11

# --> Video Opening <--

source = DirectShowSource("Amazon_720.wmv")

video = DirectShowSource("Amazon_720.wmv")
video = Lanczos4Resize( video, 720, 576)
video = BilinearResize( video, 1280, 720)

# --> SSIM analysis <--
return SSIM(source,video,"results.csv","SSIM.txt",lumimask=true)

Playing simulation of q2 MPEG4 ASP 720p encoding from 576p DVD on 720p screen : SSIM = 86.60

# --> Video Opening <--

source = DirectShowSource("Amazon_720.wmv")

video = DirectShowSource("XviD-720p-q2.avi")

# --> SSIM analysis <--
return SSIM( source, video, "results.csv", "SSIM.txt", lumimask=true)

Playing simulation of lossless 720p encoding from 576p DVD on 720p screen : SSIM = 94.02

# --> Video Opening <--

source = DirectShowSource("Amazon_720.wmv")

video = DirectShowSource("Amazon_720.wmv")
video = Lanczos4Resize( video, 720, 576)
video = Lanczos4Resize( video, 1280, 720)

# --> SSIM analysis <--
return SSIM(source,video,"results.csv","SSIM.txt",lumimask=true)


q2 576p encoding vs q3 720p encoding

Playing simulation of q2 MPEG4 ASP 576p encoding from 576p DVD on 720p screen : SSIM = 79.70

# --> Video Opening <--

source = DirectShowSource("Amazon_720.wmv")

video = DirectShowSource("XviD-576p-q2.avi")
video = BilinearResize( video, 1280, 720)

# --> SSIM analysis <--
return SSIM(source,video,"results.csv","SSIM.txt",lumimask=true)

Playing simulation of q3 MPEG4 ASP 720p encoding from 576p DVD on 720p screen : SSIM = 83.17

# --> Video Opening <--

source = DirectShowSource("Amazon_720.wmv")

video = DirectShowSource("XviD-720p-q3.avi")

# --> SSIM analysis <--
return SSIM(source,video,"results.csv","SSIM.txt",lumimask=true)

[SeeMoreDigital]

Hi bleo,

I don't think I'm missing the point...

My issues PAR has never been with regard to the ITU specification either... it's just that "others" keep bringing it up.

It sounds like you and I want the same thing. Maybe others do to!

I generate my DVD back-ups at 1:1, whether they be PAL or NTSC, 16:9 anamorphic widescreen or 4:3 standard.

After generating the encodes I often use Moitah's MPEG4 Modifier to apply/alter the AR signalling within the Mpeg4 video stream. And it was during this procedure I noticed the following anomaly...

When I applied "16:9 PAL" PAR signalling, my software media players reported this: -

http://img121.exs.cx/img121/8412/169PAL_PAR_setting.png


But when I applied "16:9" DAR (ie: 64:45 PAR) signalling, my software media players reported this: -

http://img124.exs.cx/img124/5779/169_DAR_setting2.png


When the encodes were visually checked against the source (using a verity of analogue and digital computer monitors and TV's). The pixel positions of encode with "DAR signalling" exactly matched the source... unlike the encode with "16:9 PAL PAR signalling".

This is the principal reason why I select DAR or use my own "custom" PAR settings.

It has never been my intension to cause arguments between forum members, only report my observations and invite other to confirm the same.


Cheers

[Sagittaire]

1024*576 1049*576 1048*576 or 1047*576 ... that doesn't change anything for my eyes and I prefer 1024*576 for mod16 and screen compatibility ... lol

but for ITU-R 601

328 / 225 PAL 16/9
82 / 75 PAL 4/3
40 / 33 NTSC 16/9
10 / 11 NTSC 4/3

[bleo]

My generalised observations are:

- If you want your encodes to have the same aspect ratio as the source viewed with a software DVD player, then encode at 1024x576 or with the generic PAR of 64/45.

- If you want your encodes to have round circles and square pixels then, in most cases, encode at 1048x576 or with the DV PAR of 16/11.

As always, the choice is yours! I am happy to help with any resizing calculations.

[SeeMoreDigital]

Originally posted by bleo
My generalised observations are:

- If you want your encodes to have the same aspect ratio as the source viewed with a software DVD player, then encode at 1024x576 or with the generic PAR of 64/45.

- If you want your encodes to have round circles and square pixels then, in most cases, encode at 1048x576 or with the DV PAR of 16/11.

As always, the choice is yours! I am happy to help with any resizing calculations. I totally agree with you...

I've also conducted tests for viewing via hardware/stand-alone players and ...

- If you want your encodes to have the same aspect ratio as the source viewed with a "stand-alone" Mpeg2/4 DVD player, then encode at 720x576 with the generic PAR of 64/45 (or 16:9 DAR).

Hardware players used: -
Sigma Xcard - with PAR/DAR signalling detection in MP4,
MK chip-set players - with PAR/DAR signalling detection in AVI
I-O DATA AVLP2/DVDLA and DVDG - with PAR/DAR signalling detection in MP4.


Cheers

[bleo]

@Sagittaire
What are you trying to show with your tests?

I might be interested in upsizing for HD displays...

[edit:]
Conclusions I have drawn from your tests so far: (some of them are obvious)
- Lanczos4 upsizing gives higher SSIM than bilinear upsizing
- Encoding to xvid q2 reduces SSIM
- Lanczos4 upsizing before encoding gives a higher SSIM than encoding at standard resolution at the same bitrate and bilinear upsizing on playback

[SeeMoreDigital]

Originally posted by bleo
@Sagittaire
What are you trying to show with your tests?

I might be interested in upsizing for HD displays... Don't waste your time creating upsized encodes from DVD sources... Your screen should be able do this for you, automatically.... mine does!


Cheers

[bleo]

Of course it upsizes automatically on playback, but like what was discussed earlier in the thread, I am interested in more control over the upsizing method. Even DVDs upsized to HD using whatever the default method of PowerDVD is, look rather disappointing :p

Also, I don't necessarily intend to upsize before encoding either. Perhaps a simple avisynth script for playback is all that is required...

[Didée]

Originally posted by SeeMoreDigital
Your screen should be able do this for you, automatically.... mine does! Well ... I could let some mates do the backup'ing of my DVDs. But how comes I prefer to do that by myself ...

Don't waste your time creating upsized encodes Okay, the following is a DVB source, not DVD. And okay, it shows (yet) some artefacts, since it's *WIP* of a certain processing chain, not finished yet. But nonetheless: imagine you have a source looking

like this:
http://img67.exs.cx/img67/3297/source_orig.jpg

Then a realtime upsize, be it SW or HW, will look

similar to this:
http://img67.exs.cx/img67/9378/source_upsized.jpg
- not more, not less.


But when blowing your recommendation in the wind, the travelling goes in

this direction:
http://img67.exs.cx/img67/5732/source_processed.jpg

(Save the last two to HD and switch forth-back)


Investing a little imagination on having the leftover artefacting solved (it's mainly a "only-8-bit" precision problem), you should get my point.

Once more: upsizing is not about "more detail", but about "better representation" of what is there.

Ever heard anyone saying "Bodybuilding is useless, because the mere number of muscles will not increase." ? No you haven't, because that's a nonsense argument, isn't it.
- But then, building up muscles is a slow process ...

[Teegedeck]

I can think of only one argument against upsizing to a square pixel aspect before encoding; but it is a rather decisive one: The question one would have to ask oneself before each encoding-session would be: Is the increased bitrate/space that is needed to encode the increased amount of pixels a meaningful investment or not? Whenever upsizing before encoding means encoding at a higher quantizer (because of your personal reasons for assigning this much, and not more, diskspace to a specific encode) the advantage of the more exact resizing might get lost, or so I think, and then the answer to that question would have to be 'no'.

Or do I go wrong here?

Edit: So it would seem to me that upsizing before encoding would be apt if you want to, say, increase the quality of your DVDs by upsizing with avisynth instead of letting the display driver do this, and for that aim are willing to sacrifice the ability to store more than one movie onto 1 DVD? Valid point, though not my cup of tea.

[708145]

Hmm. To me the last picture looks more artificial than the second.
On a CRT the effect was neglible but on a LCD it's very pronounced.

Are you sure you only changed the resizing method? Or is there sharpening involved?

bis besser,
Tobias

[Soulhunter]

Originally posted by 708145

Are you sure you only changed the resizing method? Or is there sharpening involved?

I think he used iip... ;)


Bye

[SeeMoreDigital]

Hi Didée,

While your enlarged and "processed" frame does at first site look impressive... you can't add more detail to an image if it isn't there to begin with - Only the illusion of more detail.

I have a 42" high-def screen (with component video, DVI and VGA inputs) and have done many tests but have found there's only so far you can go when it comes to increasing the vertical resolution. But like I say, whether it's worth the hassle is debatable.

All I need do to get the same effect as your "processed (960x544) image" is, increase the sharpness setting of my screen. Which is something I do (from time to time) when watching DVD's.


But I'm curious though...

The source you provided is from Enterprise, which is transmitted in some parts of Euroland at 1.777:1 using 720x576 pixels with an 1.777:1 anamorphic AR.... So how come your source image is only 704x432? Which after processing, has created a frame of 960x544. And therefor contains less vertical resolution than the DVB source!


Cheers

[PiXuS]

Originally posted by SeeMoreDigital
While the enlarged frame "does" at first site look impressive... you can't add more detail to an image if it isn't there to begin with - Only the illusion of more detail.



Hmmmmm... didn't Didée specifically said Once more: upsizing is not about "more detail", but about "better representation" of what is there. ???

Though... I believe Soulhunter is right... Didée didn't "simply" upsize using a more precise upsizer, but he processed the picture with iiP or LimitedSharpen or whatever Didée is brewing these days. :)

Personnaly, I cannot afford the amount of bits such a big picture demands. I prefer to encode in anamorphic and tell ffdshow to resize using Lanczos... when CPU's will be fast enough, I will use something other than Sharpen to do real time sharpening through ffdshow.

just my 2 cents...

[SeeMoreDigital]

Okay, I've ammended the first line of my previous post to suit. As I'm fully aware that Didée did not simply "enlarge" the image :eek:

Which is why I mentioned what I would do to obtain the same effect, when viewing via my HD screen!


Cheers

[Didée]

Oh-oh, what have I done :D

Firstly: No, of course that's not only a "simple" upsize. And it wasn't iiP, either. Instead I was toying around with the new features of LimitedSharpen's next version, and especially with the possibilities of doing uber-enhancement by chaining several insances of it. In fact, the last picture was done by 3 (three) serial instances of LS(), the first two with insane strength settings, and in-between denoising. You might try to put any other sharpener in a three-fold way ...


PiXuS: Dunno how it comes, but usually I'm encoding my upsized & sharpened clips with moderate bitrates (1000 to <3000, depending), at rather low quantizers ...


SMD:

"German" DVB isn't very impressive. And when something happens to get transmitted anamorphic, then we make a BIG red cross in the calendar :/
And the 960*544 is "1024*576, overscan-reduced" ... but that doesn't really matter here. It could be any resolution, and it was only testing-around.

Then, with all respect, I think you severely underestimate how poor those ENT sources look when they're in motion. That is, even on a "normal" 100Hz TV-set. I suppose you'd be very surprised about how really bad they'd look on your hi-def screen! (Especially since they're crappy field-blended, oh my.)

And, again with all respect ;) , I *absolutely* doubt that you can get THAT amount of sharpening by just increasing the sharpness of your screen. I know what insane settings I had to use, how badly the pre-present EE became over-prominent, and how much noise there had to be reduced to be able to enhance anything at all - but without loosing those details I wanted to get back ...

Also, consider the case that that source shall be re-encoded (although the source isn't good quality, it sucks too much bitrate: ~4500 kbps - big waste of space to store a poor looking source).
And if you "simply" re-encode such things, then you'll loose pretty much of those fine details that you almost can not *see* at all in that original source. Look at Jolene Blalock's lips. Look at the hair. Look at those small dots (how's the english word?) in her face, and at the fine texture at her neck's skin (latter is invisible in the source). You know very well how much of all of that you'll loose through fast&easy re-encoding: the biggest part of it, unless you go for (again) insanely high bitrates, to preserve those small, almost invisible details ... MPEG isn't suited that well for preserving such weak information.


Hey, I didn't say "this is what it has to look like". I only wanted to show how much detail (even "invisible" one) can be squeezed out, as long as it is there. Because after re-encoding, it's mostly gone and lost for sure.


The lame go to sharereactor & Co. -- I instead prefer trying to be creative, perhaps yielding to something usable and useful, at the end of the day. Let the lame be lame ... they're lame ;)

[SeeMoreDigital]

Originally posted by Didée
Then, with all respect, I think you severely underestimate how poor those ENT sources look when they're in motion. That is, even on a "normal" 100Hz TV-set. I suppose you'd be very surprised about how really bad they'd look on your hi-def screen! (Especially since they're crappy field-blended, oh my.) Actually, no I'm not surprised, because DVB-S and T is much the same here in the UK.

As I write this, I'm watching channel "five" via DVB-T on a Hitachi 1024x1024 high-def 42" plasma... and it looks bloody dreadful. If I switch over to analogue, it looks much better.... what a pisser!

But thanks for explaining a bit more how you generated your encode.


Cheers

[Sagittaire]

if hardware resize is ~ bilinear
if software resize (pre-process) is Lanczos

for metric on 720p screen:
1) 576p DVD source is better than 720p MPEG4 q2 encoding
2) 720p Lossless encoding is better than DVD source
3) 720p MPEG4 q3 encoding is better than 576p MPEG4 q2 encoding

[lamer_de]

My point, which I already rised on page 1 still remains: What if you process your image with those sharpeners, but keep the original resolution and encode at that res and then upsize during playback? I heavily doubt you can spot a significant difference during playback then.

Oh, and the sharpening you used reminds me heavily of xsharpen artifacts, especially prominent around that silverish thing in the background. Yeah, you already stated it was for demonstrating purposes only, just wanted to point out that the unprocesed one looks much better to me ^_^

CU,
lamer_de

[EnergeticBence]

Just who would watch an upsized video on a TV?
Even the best of them can't reproduce the quality of a good 22" screen.
What I do ( To prevent resize quality loss ) is set 1440x960 res on the screen @96Hz ( pstrip ) so I get no resize qiality loss and the screen is in sync with the movie... :)

[Didée]

Originally posted by EnergeticBence
What I do ( To prevent resize quality loss ) is set 1440x960 res on the screen @96Hz ( pstrip ) so I get no resize qiality loss and the screen is in sync with the movie... :)
Youthful optimism. There is no "quality loss" by upsizing to non-integer multiplies of the original resolution.


source (6x zoomed) :
http://img50.exs.cx/img50/8083/4f-source_zoom6x.png

upsized 150% (3x zoomed) :
http://img50.exs.cx/img50/8365/1e-upsize150_zoom3x.png

upsized 200% (2x zoomed) :
http://img50.exs.cx/img50/7449/2f-upsize200_zoom2x.png

I'm curious what "quality loss" you see in the 2nd one?


Even the best of them can't reproduce the quality of a good 22" screen.In exchange, your good 22" screen most likely has severe problems with interlaced content ...

[SeeMoreDigital]

Originally posted by Didée
..."German" DVB isn't very impressive. And when something happens to get transmitted anamorphic, then we make a BIG red cross in the calendar :/
And the 960*544 is "1024*576, overscan-reduced" ... but that doesn't really matter here. It could be any resolution, and it was only testing-around.) Actually, after re-reading your post, I would like to understand why you had to reduce an 16:9 anamorphic image containing 576 vertical pixels, down to 432 vertical pixels, prior to re-encoding!

Here's a typical UK DVB-T/S, 16:9 anamorphic 720x576 image from BBC News, without DAR correction: -

http://img122.exs.cx/img122/4623/0f-SMD_DVB-T_capture.th.png (http://img122.exs.cx/my.php?loc=img122&image=0f-SMD_DVB-T_capture.png)

And the same image again, with DAR correction: -

http://img122.exs.cx/img122/7453/37-SMD_DVB-T_capture.th.png (http://img122.exs.cx/my.php?loc=img122&image=37-SMD_DVB-T_capture.png)

As you can see, there's very little matte to crop away!

Are you saying, your DVB-T/S 16:9 "anamorphic" broadcasts are not presented like this.... I find this strange because I'm watching on right now!

[Didée]

In case I wasn't clear enough:

99.9% of all digital broadcasts in Germany are NOT anamorphic. So I did not reduce anything: we just don't get any more than that.

[SeeMoreDigital]

Thanks for clearing that up Didée.

I have to say, when watching "analogue" German TV via I've always been impressed to see content shown at it's correct "theatrical" aspect ratio.

With this in mind, I find it quite bizarre that more German "digital" TV stations don't transmit, what is obviously anamorphic content, correctly.

Earlier today, I watched a couple of "local" stations transmitting anamorphic content.... shame it wasn't something worth watching.


Cheers

[SirCanealot]

Originally posted by lamer_de
My point, which I already rised on page 1 still remains: What if you process your image with those sharpeners, but keep the original resolution and encode at that res and then upsize during playback? I heavily doubt you can spot a significant difference during playback then.



Because when you upscale in post-processing you upscale everything.
You upscalle everything. Everything. Details... and artifacts... :)

For example (based in NO facts, only logic :P):

Encode to 640x480 ; macro block = 1 pixel big
Upscale same encode to 1024x768 ; macro block = 4 pixels big (again, probably wrong, but you get the point)
Upscale source to 1024x768 before encoding ; end result - macro block = 1 pixel big

Upscalling after encoding = making blocking/etc bigger.
Heh, take a source, and encode it to say... Quant6.
Encode that source to 320x240 and 640x480. When you zoom the 320x240 version to fullscreen, all the blocking has been made 2x as large than when it was 640x480!

If you're upscalling before XVid (or whatever) touches it, of course it's going to look better.
I use post-processed upscalling via FFDShow a lot when I view anime. There's LOTS of limiations. Upscalling a shitty fansub source to 1280x960 only makes it look SHIT. The source needs to be perfect to look good when using post-processed upscalling.
Of course, assuming we're all "perfect god encoders", :), then upscalling after encoding can help add quality to an encode. However, it's still not as good as upscalling before encoding. End of story.

[HairDude]

To quickly come back to the original post...

"Should I encode at 720x544 (which is the same as the DVD)"

Wouldn't you be upsizing anyway if you use cropping (be it auto or manual) with let's say Fairuse Wizard? That normally always manages to find a few lines to crop off, even if it is a 16:9 Anamorphic image.

http://homepage.ntlworld.com/jontybigwang/fairusewizardfriday.png

http://homepage.ntlworld.com/jontybigwang/fairusewizardfridaycloseup.png

-
HairDude