Upscaled video - back to original resolution

[Stormborec]

I have a video file 1280×720, that was most likely created from DVD. The upscaled video is blurry, but quite detailed and high bitrate. Is there any way to recreate the original DVD?

[feisty2]

dither_resize16 (720,480,kernel="bilinear",invks=true)

[Stormborec]

Now the video has twice height and in the bottom half is color noise ...

Perhaps something like this could be usefull?:

spline36resize(960,540)
lsfmod(defaults="slow",strength=50,ss_x=1,ss_y=1)
spline36resize(640,360)

I've also tried debilinear(640,360), but the previous seems to be better ...

[feisty2]

full version:

dither_convert_8_to_16 ()
dither_resize16 (640,360,kernel="bilinear",invks=true)
ditherpost (mode=6)

[feisty2]

try kernel="bicubic" if the result looks oversharped

[Stormborec]

Thanks

Is it this package?: http://forum.doom9.org/showthread.php?p=1386559#post1386559

I'm not sure, if I true understand what it do. It computes in 16-bit depth and adds some grain and removes banding?

[TheSkiller]

Yes, that's the package.


The script posted by feisty2 just resizes in 16 Bit and dithers back to 8 Bit. Pretty basic (no debanding or adding grain)

I doubt you will notice a difference compared to a standard resize in 8 Bit though. I personally can't tell a difference between 16 Bit and 8 Bit resized videos. 16 Bit resizes are nice if you want to do gamma-corrected resizing but for standard "gamma-un-aware" resizing 8 Bit is totally sufficient imo.

[cretindesalpes]

To invert an upscale known to be bicubic or bilinear, you can use debilinear or debicubic (see here (http://rgb.chromashift.org/)).

[feisty2]

16bit is not the main point here
it's invks
it reverses upscaled image if you know the original resolution and upscale algorithm

[Stormborec]

Thank you very much.

I've tried the script with dither and also the debilinear and the output is for my eyes the same ... Perhaps it is due to I dont'n know the original resolution ... (sorry my english)
Maybe I'll deside not to downsize the video, I'll better same realtime sharpening ... My idea is - if the video is just upsized, I could use lsfmod without supersampling (ss_x=1, ss_y=1). But I'm not sure, if the supersampling is making the job just during the downsizing phase ... (in defaults="slow" it upsizes 1,5×, then it sharpens and then it downsizes 1,5×).
The video is theoretically allready upsized from 720×576(or 480) to 1280×720. What is then the correct way?

[Stormborec]

I wrote it unintelligibly ...

the question should be: how to sharpen video , which has been already upscaled

[SeeMoreDigital]

dither_resize16 (720,480,kernel="bilinear",invks=true)The source could be PAL (720x576). Stormborec, what is it's fps speed?

[Stormborec]

25 FPS, but it's just TV serial rip. It isn't worth many hours to process ...

[SeeMoreDigital]

25 FPS, but it's just TV serial rip. It isn't worth many hours to process ...Okay then. So originally it was most likely a PAL source encoded using 720x576 pixels...

[Stormborec]

I just interested how it would look after appropriate treatment...
SG1 ...
There is also some color shift - to red (perhaps wrong color space)

http://imgur.com/z4xkepN
http://imgur.com/8HrPrlh

In the video is only litle haloing and few of noise, but I'm not sure, which resizer was used, and perhaps it would be better to do only litle sharpening, and perhaps something like colormatrix ...

[Stormborec]

Maybe something like this could take sense:

debilinear(720,576)
ColorMatrix(mode="Rec.709->Rec.601")
spline36resize(768,432)

[colours]

There's nothing wrong with using ColorMatrix, but it'd be better to do the conversion with 16-bit intermediate precision (check Dither's documentation for an example on how to do that) or to flag your resulting encode as being BT.709. Colourspace conversion is one of those things that really benefit from having higher precision.

Also, PAL DVDs have active picture size only 702×576, not 720×576. Not that a 2.5% difference in aspect ratio will matter much, though.

[AzraelNewtype]

Thank you very much.

I've tried the script with dither and also the debilinear and the output is for my eyes the same ... Perhaps it is due to I dont'n know the original resolution ... (sorry my english)


That's what shrinker, from the same site, is for. Take a screenshot, save it as a 24 bit bmp, run shrinker.exe on it. It should give you a much better idea of what the original resolution is. Any of the upscale inversion tools are going to give fairly questionable results if not you're not actually using them to invert, but just as generic downscalers.

[Stormborec]

What do you mean - "shrinker"?

[colours]

http://rgb.chromashift.org/shrinker%20r1.zip

In my experience sufficiently few things are bilinear-upscaled that this tool pretty much always generates irrelevant results, but it can't hurt to try.

[Stormborec]

So, shrinker detected resolution 1120×625 of the first and 1190×661 of the second picture from the same video. If I understand - no bilinear ...