DV Info Net - Maximizing HD to SD Quality

DV Info Net (https://www.dvinfo.net/forum/)

- What Happens in Vegas... (https://www.dvinfo.net/forum/what-happens-vegas/)

- - Maximizing HD to SD Quality (https://www.dvinfo.net/forum/what-happens-vegas/271329-maximizing-hd-sd-quality.html)

Re: Maximizing HD to SD Quality

Quote:

Originally Posted by Garrett Low (Post 1629115)

Brian, to get back to your quality issue, what algorithm were you using when you use VD? The default is not Lanczos3. I believe it defaults to Bicubic. You might try to check that and see if you still see a difference. Also check to see that the resolution and PAR correct for the VD encoding so that Vegas is not resizing the video when you create your widescreen mpg file. If you're using the same algorithm and same compression codec, it should look the same.

I used the Lanczos3, I always followed Perrone's lead since he understands this stuff so well. I think I did 720x480, but I didn't give VD a fair shot and most likely didn't get near the results VD is capable of and didn't have time to buckle down with it , also I didn't like the interface for some reason. But with TMPG, right out of the gate, first attempt with the trial version and BANG, dramatic improvement. So I pulled the trigger and pulled out the Visa Card. Don't tell my wife.

For sure though, VD has crazy power for a free program.

Brian

Re: Maximizing HD to SD Quality

Hi

Just found an AVISynth script (via someone else that found it at Doom9) that seems to be doing a fantastic job at resizing HD to SD, and does it without any de-interlacing required avoiding some nasty artefacts that can introduce sometimes.

So thought I would share, here is the resizing script:

Code:



# Converts 1080i to 576i 

clip = AVISource("Source.avi").AssumeTff()



IResize(clip,704,576)

function IResize(clip Clip, int NewWidth, int NewHeight) {

  

  clip.SeparateFields() 

  Shift=(GetParity() ? -0.25 : 0.25) * (Height()/Float(NewHeight/2)-1.0)

  E  = SelectEven().Spline36resize(NewWidth, NewHeight/2, 0,    Shift)

  O  = SelectOdd( ).Spline36resize(NewWidth, NewHeight/2, 0,   -Shift)

  Ec = SelectEven().Spline36Resize(NewWidth, NewHeight/2, 0,  2*Shift)

  Oc = SelectOdd( ).Spline36Resize(NewWidth, NewHeight/2, 0, -2*shift)

  Interleave(E, O)

  IsYV12() ? MergeChroma(Interleave(Ec, Oc)) : Last

  Weave()

}

I use it by first exporting the finished project using the Lagarith codec set to RGB at the native HD resolution. Then modify the script for the correct file name and open the AVISynth script in Virtual dub and set it to fast recompress, with the video compression set to Lagarith again, this gets out a new AVI resized using the script, that AVI file can then be dragged back to Vegas and then exported using the DVD templates.

If anyone has any test footage they would like me to try this with and post back up somewhere more than happy to. I found this virtually gets rid of interlacing artefacts and movement and camera pans are also clean, probably due to no de-interlacing. Not absolutely as sharp as some other methods, but then it compresses easier and looks great to watch.

Regards

Phil

Re: Maximizing HD to SD Quality

Quote:

Originally Posted by Phil Lee (Post 1647742)

... without any de-interlacing required avoiding some nasty artefacts that can introduce sometimes.

I don't think you can resize without de-interlacing ... resizers typically expect pixels to be fairly evenly distributed and in interlaced material you miss every other line...

Have been using SD so far to avoid the problems with resizing interlaced video, but first tests with nnedi and lanczos look promising.

Re: Maximizing HD to SD Quality

Hi

I thought the same. In the above example, it is separating out each field and resizing the field to create a new field at SD resolutions, and seems to be resizing in such away to avoid nasty resizing artefacts.

It really does work very well, perhaps not as sharp as some methods that have a de-interlacing step, but I always find pans across completed scenes becomes quite messy with a big drop in detail due to the de-interlacing, but this isn't the case with this method avoiding de-interlacing, and you are not messing around with the temporal detail as you are by interpolating up to 50 de-interlaced frames to resize and derive new fields from.

Regards

Phil