Compressor 4-improve the video quality

[Mållgan]

In this article

http://www.larryjordan.biz/compressor-up-res/

mr Larry Jordan describes how to improve the video quality using the software Compressor 4 (or 3). This software is an Apple-software. Does anybody know if there is a similar option using PC-sofwares, for example Adobe Premiere Elements and Magix Movie Edit Pro MX. I have tried the "Super Resolution"-software (a Virtual Dub-plugin) but the result is zero (as far as I can notice) and using Super Resolution demands that you double, at least, the input video pixel size.

[Hoser Rob]

I had a look and wasn't that impressed at the final results. At least he's honest enough to say it won't always work that well, but I think that's being generous.

There's just no way you can make 480p sd as good as 11080p (or even 720p IMO) hd. You can't put in information that isn't there. Your eyes may be able to tell what it should look like but software can't.

I certainly wouldn't consider it worth the encoding time required.

[Mållgan]

As mr Jordan says: "SD video doesn’t contain enough pixels to fill an HD image. An NTSC SD DV image contains 345,600 pixels, while a 1080 HD image contains 2,073,600. Talk about scraping too little butter over too much bread".

BUT the Compressor 4-software does not work that way as far as I can see. It seems to work similar to the way the "Super Resolution" Virtual Dub-plugin works.

[jagabo]

No upscaler really restores the original detail. At best, they attempt to upscale while keeping sharp edges but not producing jaggies and buzzing edges when in motion. Super Resolution techniques use multiple frames to better analyze how the upscaled image should look. They work best when there is a tiny bit of motion between successive frames. Not so well with still shots or moderate to high motion shots.

Ifognition cherry picks their shots and vastly exaggerates the before and after images at their web site. They gratuitously show nearest neighbor upscales (the worst possible method) to their upscaling method. If you haven't seen it, you might find their detailed comparison of several different algorithms useful:

http://www.infognition.com/articles/video_resize_shootout.html

[Mållgan]

I have tried the Infognition Super Resolution software using SD-vhs-4:3-files as input. My experience: useless

[jagabo]

Try something like:

nnedi3_rpow2(2)
Sharpen(0.3)
FlipVertical().FlipHorizontal()
nnedi3_rpow2(2)
Sharpen(0.3)
FlipVertical().FlipHorizontal()

That's a 4x scaling that looks better than a single nnedi3_rpow(4). needi3 tends to shift the image up and left. The Flips help keep it centered. Of course, you want to remove blocking and DCT ringing artifacts and other noise before scaling.

[theewizard]

what encoder is that script code for
will that work with avisynth in avs2dvd

providing of course the final ouput does not exceed dvd frame size standards

[poisondeathray]

(as far as I can notice) and using Super Resolution demands that you double, at least, the input video pixel size.

providing of course the final ouput does not exceed dvd frame size standards

Maybe there is some misunderstanding -

Guys, that's what those articles , programs , scripts, mentioned in this thread are used for : Upscaling. ie Increasing the frame dimensions. Upscaling to HD of course won't be DVD complaint

You can still use SR techniques by multiples of 2 (2x, 4x, etc...), and then downscale to your final resolution. If input = output resolution, then of course there will be no benefit

[theewizard]

super resolution in vlc , really doesnt interest me
i've seen those 'results' reports

the quoted nee3 doublr flip double upscaling interest me

in converting some old avi rips processed for vcd and mobile
to bring up the res to DVD size, IF the results is better than a straight lacross resize

that is my question , how does it compare , and will it work in avisynth

[poisondeathray]

in converting some old avi rips processed for vcd and mobile
to bring up the res to DVD size, IF the results is better than a straight lacross resize

It will be better than a straight lanzos resize, there will be less aliasing. On something like cartoons, it will be very noticeable, but on other types of content, live action, not as much

[poisondeathray]

needi3 tends to shift the image up and left. The Flips help keep it centered.

You can use "cshift" parameter of nnedi3_rpow to prevent shifting

Ifognition cherry picks their shots and vastly exaggerates the before and after images at their web site. They gratuitously show nearest neighbor upscales (the worst possible method) to their upscaling method. If you haven't seen it, you might find their detailed comparison of several different algorithms useful:

BTW many the "PSNR" values that Infognition present are misleading because of the shifting that isn't corrected for in their testing. So of course upscaled algorithms that shift will create lower PNSR values...

[jagabo]

needi3 tends to shift the image up and left. The Flips help keep it centered.

You can use "cshift" parameter of nnedi3_rpow to prevent shifting

Oh, thanks. I thought that was just for getting non power of 2 sizes.