Up-sampling 480i to 720i

Surge wrote on 6/17/2003, 4:51 PM

I was wondering if anyone had any good results or thoughts about rendering DV-NTSC source material at higher resolutions than 720x480. Say, take a 720x480 clip and render it at 1280x720. I realize that data would need to be interpolated, but could it yield a nice picture? Garbage in equals garbage out, but if a nice source 720x480 image exists, will Vegas up convert it nicely to this HD resolution. Does Vegas utilize line doubling to achieve this? Has anyone tried this, and if so anyone have any guidelines on the topic.

Comments

John_Cline wrote on 6/17/2003, 5:08 PM

First of all, 1280x720 HiDef is progressive, not interlaced. Secondly, 1280x720 and 1920x1080 are based on square pixels and not rectangular pixels like 720x480 DV.

Vegas would have to de-interlace the DV footage, which would negatively impact the vertical resolution, then it would need to convert the footage from the 4 x 3 aspect ratio based on rectangular pixels (which is actually 1.5:1 instaed of 1.333:1) to the 16 x 9 aspect ratio based on square pixels. Plus, like enlarging any other raster-based image, Vegas will have to create information that was never there to begin with.

Nevertheless, I'm sure Vegas can handle this pretty much as well as any other application.

John

RBartlett wrote on 6/18/2003, 2:14 AM

Another job for the video bus with the supersampling envelope set!

1080i might suit upsampling of interlaced SD 59.94i NTSC footage.
You'll fid it a lengthy process, but some early HD broadcasters are using silicon in rackmount kit to do this same upsample, so you might get better results than them by waiting within Vegas!

720p would suit progressive media better.

If at the end of this you plan to make an MPEG-2 or a WMV9 with a bitrate somewhere in the order of 10Mbit/sec, then don't bother at all. You need to think double, treble or quadruple if compressing.

Nice to have the tools to hand, even if they are bleeding edge.

Spot|DSE wrote on 6/18/2003, 11:32 PM

Lengthy indeed, but there is a tutorial on the SMG site for taking exceptionally low resolution media, upsampling using the SuperSampler. I've also done this with Supersample going up to 1080i. Looks pretty darn good on my projection system at home. (1080i)
Warning. Not for weak of heart or cheap hard drives. Takes a long time, system gets pretty hot.
I added .01 of blur to the upsample just to clear the edges a little.

rebel44 wrote on 6/19/2003, 12:02 AM

I have a hard time to believe that you can blow the image to that resolution and have good picture of it. From 480 to 720 that is 50% more. You can easy see it in any graphic program when you zoom image by 50%.

jbl375 wrote on 6/19/2003, 1:00 AM

How to go to SMG site?

RBartlett wrote on 6/19/2003, 2:04 AM

Quickly register for free then enter....click here.

TorS wrote on 6/19/2003, 6:01 AM

There has been some confusion over that particular tutorial (See the thread: What is supersampling?)
because some people have tried to reproduce it and been getting results nowhere near the rather striking "before and after" images of the tut.
Tor

farss wrote on 6/19/2003, 8:14 AM

I've done a very small test on this, mostly using generated media and HiRes stills and IMHO it looks every bit as good as it should. If your starting from DV and upscaling to HD needless to say no miracles are going to happen but if you are planning on projecting the image with a video projector then most of them will be doing the same process and I'd imagine VV taking its time to do it versus the projector doing it in real time has got to give better results.

I notice some of the higher end projectors now have DVI inputs, it would be interesting to produce some 720 footage and try it straight from a PC into the projector using the DVI interface.

mikkie wrote on 6/19/2003, 9:24 AM

FWIW, do your own tests and all, supersampling in what tests I've run is not designed to improve basic video. It does create more frames as tweeners, interpolating the image data present in the original frames at both ends of the spread. While I could of course be overlooking something, I just don't see where having extra frames, less accurate ones at that as they are derived from interpolation, could enhance resampling the data in one pre-existing frame and basically blowing it up, the same way one enlarges a picture in P/Shop.

I think it *might* be possible to come up with a scheme where you split the fields into frames (in V/Dub for example), import the roughly 60fps footage into Vegas, and see some improvement going to 29.97p from there using supersampling, but it would have to be tested.

Realizing it's not a popular sentiment here, IMO V/Dub still does a better job, quite a bit faster when it comes to resizing, so might want to try that as well.