Fixing aliasing/combing in PAL progressive help

[spiritgumm]

Not sure why they released it this way, but hoping it can be decombed or aliasing improved before I change to ntsc specs.

[manono]

It's all aliasing, not combing, caused by them making it 4:3 and not 16:9. There's not enough resolution to resolve fine lines.

When anti-aliasing it's a fine line between getting rid of the aliasing and losing too much other detail, so you have to decide for yourself when enough is enough. Or apply it just on the parts needing it and leave the rest alone. This would have been much worse in a 4:3 NTSC DVD. Anyway, you can try:

QTGMC( Preset="Slower", InputType=2, ProgSADMask=12.0 )

And decide if it's good enough for you. The work that can be done improving progressive sources such as this is well explained in the QTGMC doc.

[spiritgumm]

Thanks, I'll give it a shot.
When using separatefields, it looks like the duplicate field in each pair isnt an exact duplicate so I was wondering if it was possible to output the first field with a duplicate, omitting the original 2nd field? Would the result look better?

[johnmeyer]

When using separatefields, it looks like the duplicate field in each pair isnt an exact duplicate so I was wondering if it was possible to output the first field with a duplicate, omitting the original 2nd field? Would the result look better?

I'm not sure what you mean. I looked at the video, field-by-field, and it looks like a standard progressive PAL video. Therefore, each field pair is from the same moment in time, but from a different place in space, which is why they aren't perfectly identical, i.e., the even fields are NOT just the odd fields shifted by one scan line. So, if you were to drop one field and replace it with a neighbor, you'd be throwing away half of your vertical resolution. While that can sometimes be a good tradeoff, I don't think it helps anything here, because the aliasing from the improper resizing will not be made any better by doing that.

[spiritgumm]

I though it might eliminate the mice teeth, which is perhaps more prominent in other scenes.

[themaster1]

Try this:

main=last
main
Santiag() # reduce aliasing
filtered=last
main
mt_edge().mt_expand
mymask=last
mt_merge(main, filtered, mymask)

note: this code keeps finer details. It's better than using santiag only imo, zoom x4 and you'll see

[spiritgumm]

Try this:

main=last
main
Santiag() # reduce aliasing
filtered=last
main
mt_edge().mt_expand
mymask=last
mt_merge(main, filtered, mymask)

note: this code keeps finer details. It's better than using santiag only imo, zoom x4 and you'll see

Thanks, but it doesnt seem to do enough.

I'd still like to know how to output duplicate fields thru decimation, etc.

[manono]

I'd still like to know how to output duplicate fields thru decimation, etc.

That question, as phrased, makes no sense. If you're referring to your earlier question:

When using separatefields, it looks like the duplicate field in each pair isnt an exact duplicate so I was wondering if it was possible to output the first field with a duplicate, omitting the original 2nd field?

then:

SeparateFields()
SelectEvery(2,0,0)
#Weave()#(?)

Would the result look better?

No.

[spiritgumm]

Thanks for the responses. I used the QTMC line with some tweaking.

Inre Weave, the frames look interlaced rather than progressive, which I guess was predictable (?).

EDIT: The tweaking was this:
QTGMC( Preset="Slow", InputType=3, ProgSADMask=80.0,TR2=3 )

[LMotlow]

Thanks for the responses. I used the QTMC line with some tweaking.

Thanks for sharing the tweaks. Hm?

Inre Weave, the frames look interlaced rather than progressive, which I guess was predictable (?).

Didn't look interlaced to me. Just sloppy processing by the maker. Line twitter and busted lines don't look like interlace IMO.

Below and attached is the best I could get with the QTGMC idea. Can't rid of the damn line twitter at the start. I see Brit films doing this often. Don't know why.

MPEG2Source("VTS_01_1.demuxed.d2v")
Crop(8,60,-8,-60)
QTGMC(Preset="very Slow", InputType=2, ProgSADMask=12.0)
supaa()
Dither_convert_8_to_16()
Dither_resize16(720,480, kernel="Spline36", cplace="MPEG2")
DitherPost()
LSFMod()
AddGrainC(2.0,2.0)
Assumefps("ntsc_film")
return last

### --- Resize mod of the old supaa anti-alias function ----
FUNCTION supaa(Clip Clp,Int"SS",Bool"CP") {
OX = Clp.Width
OY = Clp.Height
SS = Default(SS,2)
CP = Default(CP,True)
Clp = Clp.IsYV12() ? Clp : Clp.ConvertToYV12()
Clp.nnedi2_rpow2(rfactor=SS,cshift="BilinearResize",qual=3).SangNom().TurnRight()
\  .SangNom().TurnLeft().Spline36Resize(OX,OY)
CP ? Last : MergeChroma(Clp)
Return(Last) }

Would be interesting to see if anyone comes up with a good solution. I see this often in retail and broadcast films. TMC prints are some of the worst offenders.

[spiritgumm]

Thanks for sharing the tweaks. Hm?

I only tweaked QTGMC line to work better with another sample, which I'll upload.
QTGMC( Preset="Slow", InputType=3, ProgSADMask=80.0,TR2=3 )

Didn't look interlaced to me. Just sloppy processing by the maker. Line twitter and busted lines don't look like interlace IMO.

Not sure if you're talking about the original clip. I was referring to using "weave" in manono's later script after separating fields.

I like the minimizing of mice teeth (and/or dot crawl) although I had to omit the "Dither" filter because dll wouldnt load. The script didnt quite work with my other sample - had some problems around the horizontal lines in the glass wall, which you'll see.