dvd to HD

[fenomeno83]

Can I get a better resolution from dvd(720x576)?For example frame from 720x576 to 1280x720!
I read this:http://www.thedeemon.com/VideoEnhancer/

[Sharktooth]

you cant restore details if they were removed.
upsizing a video is not a good idea. what you see on that site is all marketing BS... and the picture examples are indeed fakes.

[giandrea]

Can I get a better resolution from dvd(720x576)?For example frame from 720x576 to 1280x720!
I read this:http://www.thedeemon.com/VideoEnhancer/

Well, you can get a better playback quality of your PAL 720x576 video on an HD TV if you use some software scaler instead of the built in hardware scaler (your graphic card's, or your TV's).
For example MPlayer can use several software scalers, but this works during playback. Re-encoding a video at a higher resolution is something that you shouldn't do definitively!

I will give you an example command line for mplayer.
mplayer inputfile.mpg -vf scale=-2:720 -sws X
where X can be:

0 fast bilinear
1 bilinear
2 bicubic (good quality) (default)
3 experimental
4 nearest neighbor (bad quality)
5 area
6 luma bicubic / chroma bilinear
7 gauss
8 sincR
9 lanczos
10 natural bicubic spline

This will upscale your clips to 720p or 720i using the selected software scaler. You can see for yourself what is the best quality, but I recommend lanczos or bicubic spline (9 or 10).

[fenomeno83]

I understand.but is there a real better quality?

[foxyshadis]

Since most people can't play something resized to 1080p in realtime, it might make sense to encode. But it's more sensible to upscale partway and just let the video card deal with the rest, particularly if you're using a renderer that can use bicubic. If you have a high enough bitrate it actually does look much sharper; otherwise, all it does is highlight the artifacts more clearly.

Also spending 8-16 hours of encoding on a 2 hour movie you're only going to watch once a year really isn't the best use of one's time... not that it stops me, but I have I love of tweaking.

[Sharktooth]

I understand.but is there a real better quality?
no...

[KoD]

fenomeno83, if you take a small gif avatar like those from foxyshadis or Sharktooth and upsize it, do you think the upsized gif will have more detail than it originally had ?

That's the same with upsizing from 720x576 to whatever.

If you want a more elaborate explanation, study the law of thermodynamics. (interesting how everything is realted to physics, isn't it ?)

[Guest]

If you want a more elaborate explanation, study the law of thermodynamics.
Can you elaborate on that please? I don't see the connection.

[Dust Signs]

I think he means that some processes are irreversible, p.e. removing details from a higher resolution video and shrinking it ;) I think that's the connection.

Dust Signs

[giandrea]

I understand.but is there a real better quality?no...

Actually in a certain sense... yes.

If you are going to display your video on a high res display there will be upscaling at a certain point of the video chain. You can use a better scaler and get a better upscaled picture quality, but clearly you can't add back information to the picture.

Try using nearest neighbour and bicubic and see the difference, and you get what I mean...

If you need an example I can post pictures :)

[fenomeno83]

post picture please

[Sagittaire]

no...

no .... ???

1) Visual example 64*36@1024*576

Source with simple zoom:
http://multimediacom.free.fr/Video/Original.PNG

Source with high quality resize:
http://multimediacom.free.fr/Video/Interpolation.PNG

Source with high quality resize and pre-process:
http://multimediacom.free.fr/Video/PP.PNG


2) Metric example:

source A at 1920*1080
source B at 720*576

I make Resize 720*576@1920*1080 with pointresize (simple zoom) for source B and I make metric test with source A. I make Resize 720*576@1920*1080 with Lanszos or spline (high quality interpolation) for source B and I make metric test with source A. The metrics for Lanczos will always very better (by far) than pointresize.


3) Real encoding example with DVD source:

Classic DivX encoding at 640*352 (12500 Ko)
http://multimediacom.free.fr/Video/volver_DVDRipp_Classic.avi

Overesize H264 encoding at 1280*720 (12500 Ko)
http://multimediacom.free.fr/Video/volver_DVDRipp_720p.mp4

Overesize H264 HQ encoding at 1280*720 (30000 Ko)
http://multimediacom.free.fr/Video/volver_DVDRipp_720p_HQ.mp4

[SeeMoreDigital]

Quite some time ago, Soulhunter experimented at generating 1024x576 (square pixel) MPEG-4 encodes from 720x576 (non square pixel) MPEG-2/DVD sources.

The perceived image quality was actually very good. Fast panning motion sequences did seem improved. So it would appear something in the viewers brain was being deceived, on a favourable basis.

That said, and as Sharktooth already pointed out, you can't create extra detail from an image once it's been removed (despite what you see on CSI). So re-sizing 720x576 sources to 1280x720, would not be worth the effort!


Cheers

[giandrea]

I am considering a common case where I can see clearly the difference in the scaling algorithm. I won't consider 480p to 720p because most eyes won't notice the difference between the different scalers because the scaling factor is low. Instead I will consider a PAL CIF (352x288) image upscaled to 720p for an HD display.

I upscaled it with mplayer with all the software scalers.
The results are in this file (http://motoriweb.com/public/resize_comparison.zip). The number in the file names correspond to the software scaler according to this table:

0 fast bilinear
1 bilinear
2 bicubic (good quality) (default)
3 experimental
4 nearest neighbor (bad quality)
5 area
6 luma bicubic / chroma bilinear
7 gauss
8 sincR
9 lanczos
10 natural bicubic spline


You can see clearly wich is worst in quality (4.png nearest neighbor), and to my eyes the best algorithms are bicubic (2.png), lanczos (9.png) and bicubic spline (10.png).
sincR has some color distortion and strange artifacts...

[Sagittaire]

Metric order with Avisynth:
pointresize
bilinearresize
bicubicresize
splineresize
lanczosresize

lanczos4 and spline36 done very similar metric result but spline36 done little more natural result for my eyes.

[SeeMoreDigital]

Sagittaire,

With reference to your three "upscaled" images.... Where's your source image?

[Sagittaire]

Sagittaire,

With reference to your three "upscaled" images.... Where's your source image?

Well it's very simple, first image is simple 64*36@1024*576 pointresize interpolation. If you want source you make reverse 1024*576@64*36 pointresize interpolation.

[SeeMoreDigital]

Well it's very simple, first image is simple 64*36@1024*576 pointresize interpolation. If you want source you make reverse 1024*576@64*36 pointresize interpolation.You're missing the point!

You can't demonstrate the benefits of using up-scaling filters if you are not prepared to provide a source (starting point) image. People at the very least need to see a proper comparison with "before" and "after" images!

[Sharktooth]

well... on black and white materials you can interpolate and get "some" results... on more complex frames you cant get back details if they're no longer there.
all you can get is a more pleasing look... but no improvement at all in details and however the standard DVD resolution is already enough to smooth out the pixels "squareness"...

examples:
http://www.webalice.it/f.corriga/resizetest/werewolf_orig.jpg - orignal image
http://www.webalice.it/f.corriga/resizetest/werewolf_small.jpg - sized down image
http://www.webalice.it/f.corriga/resizetest/werewolf_nn.jpg - nearest neighbor upsize
http://www.webalice.it/f.corriga/resizetest/werewolf_bl.jpg - bilinear upsize
http://www.webalice.it/f.corriga/resizetest/werewolf_bc.jpg - bicubic upsize
http://www.webalice.it/f.corriga/resizetest/werewolf_bcs.jpg - bicubic sharp upsize

none of the upszied image have more details than the small picture... and imho they dont even look better.

[scharfis_brain]

if the source is blurred (typical for DVD) you only can fake detail and edge definition using some algorithms like EDI or ELA.

but for highly aliased sources you may percieve a big gain in resolution when taking previous and futur frames into the calculation:
http://home.arcor.de/scharfis_brain/samples/schock.png

[Sharktooth]

yes, but this is not likely done in realtime...
what videoenhancer does seems just a sharp resizing... (probably lanczos4 or something).
but however you can get those kind of results only on very low resolutions sources.
DVD is already too "high-res" to spot the "smooth" upsizing effect unless you have a 200 inches TV and you're upsizing from 720x**** to 8192x****.

[Sagittaire]

With high textured image if you want:

Source 1920*1080
http://multimediacom.free.fr/Video/image2_1920x1080.PNG

Source 720*576
http://multimediacom.free.fr/Video/image2_720x576.PNG

Source 720*576@1920*1080
http://multimediacom.free.fr/Video/image2_720x576@1920x1080.PNG


1920*1080@720*576@1920*1080 is simply like make blur for original source.
HD SAP or HD TV make always 720*576@1920*720 interpolation for SD source (like DVD).
High Quality interpolation is always better for eyes.

[Sharktooth]

so, where's the gain in upsizing? it looks exactly the same to me and maybe more blurry than the original 720 source image...
the 1920 source is clearly much better than both 720 source and the upsized images.

[SeeMoreDigital]

I don't understand which image is the source?

[Sharktooth]

the source is the first image
it was sized down (image)
and then sized up (third image)

[SeeMoreDigital]

the source is the first image
it was sized down (image)
and then sized up (third image)Well, in my opinion what Sagittaire is doing is a ridiculous way of going about things. And proves absolutely nothing!

The original poster was interested in up-scaling from 720x576 to 1280x720.

Come on Sag, if you want to prove your point, show us what you can do within the parameters of fenomeno83's original question?!

[Sharktooth]

There's nothing to do unless you want to add artificial details.

[giandrea]

none of the upszied image have more details than the small picture... and imho they dont even look better.

Certainly this is obvious. But I think that the question at this point is:

if I MUST upscale the image (because I'm going to show it full screen on an HDTV panel) what is the best method of upscaling?

Because that is what our friend fenomeno83, and anyone who is going to watch a DVD on an HDTV, must do: upscale. :)
Unless you are going to see your DVD at its original resolution with black borders all around to pad the frame size to your HDTV... O_O

[Sagittaire]

so, where's the gain in upsizing? it looks exactly the same to me and maybe more blurry than the original 720 source image...
the 1920 source is clearly much better than both 720 source and the upsized images.

You are sure for that:

Source 1920*1080
http://multimediacom.free.fr/Video/image2_1920x1080.PNG

Source 720*576
http://multimediacom.free.fr/Video/image2_720x576.PNG

Source 720*576@1920*1080 simple zoom
http://multimediacom.free.fr/Video/image2_720x576@1920x1080_zoom.png

Source 720*576@1920*1080
http://multimediacom.free.fr/Video/image2_720x576@1920x1080.PNG


Well, in my opinion what Sagittaire is doing is a ridiculous way of going about things. And proves absolutely nothing!
The original poster was interested in up-scaling from 720x576 to 1280x720.
Come on Sag, if you want to prove your point, show us what you can do within the parameters of fenomeno83's original question?!

... lol

1) If you want make 720*576 encoding you must make 720*576 resizing with the master.

2) My example is simple spacial interpolation. For video like DVD, high quality SAP upscaler can use spatio-temporal interpolation with motion vector interpretation. I will make demo if you want ...

[SeeMoreDigital]

I will make demo if you want ...It's not what I want... It's what the thread poster wants!

On a personal note. I'm able to scrutinise 720x576, 1280x720 and 1920x1080 still images or video on my high-def capable display!

[Sagittaire]

The original poster was interested in up-scaling from 720x576 to 1280x720.



Can I get a better resolution from dvd(720x576)?For example frame from 720x576 to 1280x720!
I read this:http://www.thedeemon.com/VideoEnhancer/

YES ... !!!

Spatial interpolation is possible
Temporal interpolation is possible
Spatio-Temporal interpolation with MV interpretation will done really good result

[Sharktooth]

bah.. i cant see those "very good" results.
if details were removed they're no longer there. you can interpolate but you cant reconstruct what's missing. Period.
And yes, im sure.. the first image looks better. Also the 720x picture looks better than the sized up images too coz one is "blocky" (due to the nearest neighbor resizing and pixels stretching) and the other is "blurry".

[Sagittaire]

bah.. i cant see those "very good" results.

http://home.arcor.de/scharfis_brain/samples/schock.png

........... ?????


if details were removed they're no longer there. you can interpolate but you cant reconstruct what's missing. And yes, im sure.. the first image looks better. Also the 720x picture looks better than the sized up images too coz one is "blocky" (due to the nearest neighbor resizing) and the other is "blurry".

Well you see neighour resizing for SD TV (simply because there are ~720*576 pixels) and you see high quality resizing for HD TV upscaling.

Interpolation create better gradiant and done better PSNR (pixel by pixel comparison). You can use high quality pre-process like sharpenig too ...

demo with spacio-temporal interpolation in progress ...

[Sharktooth]

once again that image is not a dvd resolution image...
also interpolation cant create better PSNR coz it alters its pixels.

[scharfis_brain]

if a dvd resoluted image would be aliased that much, it would work the same.

[shon3i]

I not want to compare two different things, but will be same if i use 44khz audio and upsample to 96khz, and add some haronics above 22khz in all channels, this file will be sound much better that 44khz, but real different than real 96khz source, and much whorse.

You can postprocess, audio and video and make this things, but that is very bad.

[guada2]

[QUOTE]once again that image is not a dvd resolution image...
also interpolation cant create better PSNR coz it alters its pixels.

True.
But, i have some questions.
- upscaling is not a "good idea", but why some materials use it today? (DVD to SD or DVD to HD or SD to HD)
- what chips are more interesting for a real upscaling?
Mustang / DCDi + or Mustang / SI

Bye.

[Sagittaire]

once again that image is not a dvd resolution image...


With DVD resolution if you want:

Simple zoom 720*576@1920*1080 pointresize (X4)
http://multimediacom.free.fr/Video/image2_720x576@1920x1080_Pointx4.png

High quality 720*576@1920*1024 interpolation (X4)
http://multimediacom.free.fr/Video/image2_720x576@1920x1080_Lanczosx4.png

Original source (X4)
http://multimediacom.free.fr/Video/image2_720x576@1920x1080_sourcex4.png

Speak about that will all SAP builder like Pionner, Denon, Sony, Philips ... with high quality upscaling.


also interpolation cant create better PSNR coz it alters its pixels.

It's false ... and it's really simple to prove that. You must compare 576p@1080p Pointresize (image with simple zoom without interpolation), Bilinear, Bicubic, Spline and Lanczos with original 1080p source and PSNR will be more and more better ...

In fact for eyes quality for SD@HD interpolation is always between SD@HD resize without interpolation (point resize or simple zoom) and original HD, metric confirm always that.

[foxyshadis]

Sagittaire, would it be possible to edit those enormous images out into links, please? ;_;

Sharktooth, what you miss in stating the obvious, that no new detail is be brought out, is that there is more detail within most smaller images than bilinear can bring out, more than spline36 can bring out, and enough that it's possible to create higher resolution upsizes with near-perfect results (although I'll admit the tools we have aren't all perfect). It's possible to stretch the information further than you'd expect, though compression, denoising, and so on remove some. By considering temporal you have enormous amounts of additional information.

Given that most systems will use a bilinear upsize, which is an extremely bad way of taking advantage of the existing detail, it can be useful. Bicubic is generally barely good enough for a 720p upsize, imho, and only lanczos/spline are watchable for even higher resolutions. Since I don't actually have a 1920x1080 or 1920x1600 anything, the only way I can test is on a friend's, where the difference is astonishing. If I was going to watch something encoded well on a regular basis, like a favorite movie, I'd spend an inordinate amount of time pre-encoding it.

One of the problems with temporal information is that encoders are also doing their best to remove all temporal redundancy, whereas for upsizing the more you have, the better. So if there's heavy ringing around an edge on the first frame and it hardly moves, it'll have the same ringing until the next keyframe.

[IgorC]

I not want to compare two different things, but will be same if i use 44khz audio and upsample to 96khz, and add some haronics above 22khz in all channels, this file will be sound much better that 44khz, but real different than real 96khz source, and much whorse.

You can postprocess, audio and video and make this things, but that is very bad.
My apologize for offtopic.

Well. Resampling 44 -> 96 khz shouldn't give any gain in audio quality. Only in very strange and odd cases.
But you can postprocess audio with dithering (I don't like this word "dither". There is nothing to do with dirt and not clean stuff. What it makes it approximate better the digital sound to human audio perception). http://en.wikipedia.org/wiki/Dithering
I just try in foobar:
8 bit - ugly
8 bit + dither - very good quality boost
16 bit - normal quality
16 bit + dither - sometimes sounds better

And if you have 24 bit soundcard there is no need to dither source just enabled 24bit decoding in sets. It sounds better but you need to have a good acoustic system to able to hear improvement

[Sagittaire]

And here Spatio-temporal interpolation with Motion Vector prectiction and high quality sharpening:

Harry Potter II 720*576*25@1280*720*50
http://multimediacom.free.fr/Video/HPIV_Interpolation.mp4

Interpolation can recreate pixel and can recreate complete frame with MV analyse.

[Sharktooth]

Interpolation can recreate gradients... not details...

[IgorC]

In theory temporal interpolation can recreate details from neighbour frames basing on motion vector analys. It's what Sagi trying to say.

[Sharktooth]

if details are removed in one frame they're removed also in neighbour frames...
the neighbour frames are at the same resolution...
the more the details the more the upscaling becomes useless.
obviously if the high res source is smooth and poor of details the upscaling can become similar...
for example, get some white noise and downscale it (let's say 33%) then upscale it to the original res with the best method you can think of.
it wont be any better... it will also loose a lot of luma info...
the resulting image would be like "blurred".

[IgorC]

if details are removed in one frame they're removed also in neighbour frames...

But it doesn't mean that exactly the same details were removed from all frames. Only strange case is if all frames are identical.

[Sharktooth]

well if the original res is not able to show subpixel details (at that res), then they're lost.
there's no way to get them back.

[IgorC]

well if the original res is not able to show subpixel details (at that res), then they're lost.
there's no way to get them back.
That's true. An enourmous part of details are lost. But there is still some percent of details that can be restored from motion. In motion the objects don't loose excatly the same pixel data.
I know that's not a good sample but think in I and B frames.

[Sagittaire]

the resulting image would be like "blurred".

Simply if you compare with the 1080p original source. But 576p@1080p interpoled image will have always better perceptual quality than the 576p image with really better details level. In fact perceptual quality for high quality 576p@1080p interpolation will be always between 576p source and 1080p source.

[Sharktooth]

Yes, if you watch your movie at 5cm from the screen...

[akupenguin]

Simply if you compare with the 1080p original source. But 576p@1080p interpoled image will have always better perceptual quality than the 576p image with really better details level.
It's not possible to compare 576p with 1080p. If you're viewing the two videos under the same conditions (i.e. same viewable area), then the 576p was upscaled too. So you're just comparing one scaling algorithm with another. Maybe one of the algorithms is in software while the other is built into your video card or even the physical properties of your crt, but either way it's scaling.

So the right question to ask is: which scaling algorithm looks best, and should it be applied before or after compression. Not whether to scale.

if details are removed in one frame they're removed also in neighbour frames...
the neighbour frames are at the same resolution...
True if and only if the content is properly band-limited.
But if it's aliased (all interlaced content is, and some badly(?) made progressive content), then the limited resolution may remove different information in neighboring frames. This makes it ugly to watch with a simple scaler (even lanczos), but otoh it makes it theoretically possible to recover some of that information with a smart temporal scaler.