lanczosresize

[insane822]

I have a 1080i leno source, when using lanczosresize (which I though would just resize) it gives me progressive video (IPIP..) pattern.
When I look at just the .ts itself, all frames are interlaced.

Does lanczosresize do more than resize? Thanks.

[stickboy]

If you give a resizer interlaced footage, it's going to blend scanlines together. The resizer doesn't know it's interlaced.

If you want it to remain interlaced, you need to do something like:
SeparateFields()
LanczosResize(w, h)
Weave()(But there are better ways (http://forum.doom9.org/showthread.php?s=&threadid=74906).)

[sh0dan]

In the example above you should remember that height should be half of your final image height.

[insane822]

yeah, I'm an idiot.

So can I make sd 'progressive' material from HD footage?

Is that what i'm alrady doing, or is this a bad idea.

[Mug Funky]

the good thing about 1080i is that vertical resolution is almost moot point for SD resizing.

if it's pure interlaced, then simply separatefields.selecteven will get you a half-height progressive clip

if it's film based, then it's probably safest to telecide or smartdecimate it (or does it come in 24p in these cases? i haven't got the HDTV fever yet).

if you're mega-worried about quality, then use a good smart deinterlacer (Tdeint, Kerneldeint, etc) and lanczos it down.

one thing though - if you're watching on a standard def interlaced TV, make sure the image isn't too sharp - it'll flicker and make bizarre patterns from the alternating scan pattern. sometimes a blur(0,.7) can work wonders (and will give you a smaller encode, too).

if you're just viewing on the computer screen, then make it as sharp as you want :)

[scharfis_brain]

with interlaced scalings in a range of 20 %, that
separatefields().resize().weave() methode may be sufficient, but with extreme ratios, like HDTV to SDTV, you'll get misplaced fields, which means, the top & bottom field are aligned spatially wrong.

the top & bottom field have the spatial vertical distance of one HDTV-Pixel, but they nedd the spatial vertical distance of one SDTV pixel.

this means, one of both fields ismisalgined by at least an quarte (or half?!?) pixel.

therefore, you should go for the 'better ways' stickyboy has linked to.

maybe, your source is decimateable, so you maybe want to try my newest incarnation matchbob() in combination with fdecimate()

[Didée]

Originally posted by scharfis_brain
you maybe want to try my newest incarnation matchbob()
OMG. scharfis_brain has incarnated in an avisynth script. :D

[scharfis_brain]

oh-oh, I should really check the words before posting :scared:

I wanted to write something likee "creation'....

I haven't got enough sleep this week.

EDIT: @didee:

you should try matchbob with restore24! it returns about 3 to 6 completely WEAVED frames per second with our fieldblended stuff!

also, how are about using fdecimate with r24? does it still crash?