Diagonal broadcast noise?

[Brad]

... and ghosting and rainbowing/dot crawl, which aren't as notable to me.



No lossless capture available; my friend's recording is "only" MPEG-2 9.2Mbps CBR. I'd like to improve the DVD a little if I can. Ideas?

[sanlyn]

The video appears to have originally been progressive + telecine. It is now interlaced and hard-telecined, was processed with telecine intact, and with overly strong temporal filters -- without removing telecine, treating the vid as "interlaced". Not much you can do about the temporal image trails, but you can try inverse telecine to clean up some of the angular jaggies. That would eliminate some blended frames and restore the original 23.97 frame speed. Then use QTGMC (Input_type=1) on the progressive video to interpolate some cleaner frames. You could then re-encode for 23.97 fps and add 3:2 pulldown flags.

[jagabo]

After IVTC you might try the DeStrip filter:
https://forum.videohelp.com/threads/345350-VHS-Restoration?p=2156015&viewfull=1#post2156015

Santiag() removes a lot of the stripe noise too. You can follow that with a resize to half width then back up to full width to remove most of the rest. But you'll get a softer picture.

Mpeg2Source("vaporeon800_93-Altima.d2v", CPU=6, Info=3) 
TFM(d2v="vaporeon800_93-Altima.d2v") 
TDecimate() 
Santiag(2,2)
BilinearResize(width/2,height).BicubicResize(width, height).Sharpen(0.5,0)

Before, after:

[sanlyn]

Yes, I still have DeStripe around. The biggest problem is bad aliasing on diagonals. It looks like deinterlace using field decimation of some sort (??? That's my guess, you're much better at spotting these things than I am).

Something like nnedi2, maybe, after TIVTC? TIVTC cleaned up a lot, but not the bad jaggies.

[sanlyn]

Noit having much luck with eedi2, but I don't know it that well. TIVTC removed most of the worst of the "bursts" of jaggies, leaving 3 frames that have this effect, which I cleaned up a bit with ReplceFramesMC2. Diagonals look smoother, mainly because the ReplaceFrames softened things a bit (the effect was in 2 consecutive frames, and the camera is panning toward the left).

After TIVTC (left), then TIVTC with RemoveFramesMC2 (right). But jaggies are still on diagonals, and this is not a really good fix for the whole video.

[Attachment 17170 - Click to enlarge]

[sanlyn]

Mpeg2Source("vaporeon800_93-Altima.d2v", CPU=6, Info=3) 
TFM(d2v="vaporeon800_93-Altima.d2v") 
TDecimate() 
Santiag(2,2)
BilinearResize(width/2,height).BicubicResize(width, height).Sharpen(0.5,0)

Right. I did the same thing, but without the Resize. Will give that a try.

[manono]

I added Santiag to the script after seeing it mentioned here (but didn't see much improvement in that digital broadcast noise), but otherwise I did it slightly differently:

MPEG2Source("E:\Test\test.d2v")
Deint = QTGMC().SelectEven()#Use QTGMC to do the deinterlacing
TFM(Clip2=Deint)
TDecimate()
A=Last
B=A.Tweak(Sat=0,Coring=True)#greyscale the M.S.R.P. text
B=B.Crop(0,372,0,0)
C=Overlay(A,B,0,372)
ReplaceFramesSimple(A,C,Mappings="[455 587]")
A=Last
B=A.Crop(0,372,0,0)
B=B.FreezeFrame(487,538,516)#freeze to the best-looking M.S.R.P. text frame
C=Overlay(A,B,0,372)
ReplaceFramesSimple(A,C,Mappings="[487 538]")
QTGMC(InputType=2)#had to do this later (to smooth the residual aliasing), I think because 2 QTGMCs aren't allowed in the same script
FreezeFrame(665,713,667)#Freeze on the best text frame at the end
A=Last
B=A.Tweak(Sat=0,Coring=False)#greyscale the text at the end
C=A.Crop(288,182,-298,-188)#and stick back in the colored logo
D=Overlay(B,C,288,182)
ReplaceFramesSimple(A,D,Mappings="[650 721]")

Interestingly, while vaporeon800 was bothered most by the diagonal crap on the windows and elsewhere, I don't think I would have noticed it had he not mentioned it. Much more annoying to me was all the jumping and rainbowed text and the aliasing on the diagonal lines.

[sanlyn]

Hmm. So....deinterlace first, then IVTC the Evens only. Cool, and the "clip2" parameter is now more clear. That gets me to reading up on more of those parameters. Maybe some way to calm those wiggly letters ? They really are a nuisance, but so is the chroma shift (the shift alone makes it look like VHS, even with the VHS noise gone).

I left santiag and other AA's behind and went with NNEDI2 routines. Worked, but didn't rid of all the shimmer. But it did make the first 10 frames fade in without sawtooths. But you'd have to look for it to notice.

[poisondeathray]

Hmm. So....deinterlace first, then IVTC the Evens only. Cool, and the "clip2" parameter is now more clear. That gets me to reading up on more of those parameters.

The whole clip isn't deinterlaced ; IVTC is first - clip2 in TFM only applies deinterlace when combing is detected after the field matching according to the threshold settings

[sanlyn]

I think I know what you mean about "not deinterlaced". Check me on this. The first QTGMC does its own "bob" doubling, but some of those Even and Odd frames have previous telecine "leftovers". TFM used with Clip2 addresses the frames with combing, then TDecimate removes the dupes. Maybe I'm not stating it correctly, but I think I know the principle. (Hope so. I gotta get to bed sometime tonite).

[manono]

I think I know what you mean about "not deinterlaced". Check me on this. The first QTGMC does its own "bob" doubling, but some of those Even and Odd frames have previous telecine "leftovers". TFM used with Clip2 addresses the frames with combing, then TDecimate removes the dupes. Maybe I'm not stating it correctly, but I think I know the principle. (Hope so. I gotta get to bed sometime tonite).

pdr's right. Most of the clip is field-matched with no use of QTGMC. But rather than use the built-in TDeint when TFM doesn't field-match, it uses QTGMC as the deinterlacer. You can put on 'Display=True' to see when it's being deinterlaced. Here it is at default settings and using QTGMC. On the left is:

TFM(Display=True)
Tdecimate()

On the right is:

Deint = QTGMC().SelectEven
TFM(clip2=Deint,Display=True)
Tdecimate()

[sanlyn]

(OMG, did I really post thank-you last night and forgot to hit Submit ? Oh, brother!)

Effective explanation and example, manono. Thanks to you and pdr for clarifying. That's one of several hurdles I had getting deeper into TIVTC.

I tried manono's script, added some anti-chroma-shift, and changed the two QTGMC's from default "slow" to "medium". This kept more detail in the guy's hand but didn't seem to effect shimmer or other glitches. Going faster than "medium", ye olde original artifacts started sneaking back.

[manono]

Yeah, that's why I also ran 'QTGMC(InputType=2)' over it afterwards - to smooth it out some more. Nothing really worked with the text, which is why I choose a decent-looking frame for the the whole time it was on.

By the way, just turning off the deinterlacer works also:

TFM(PP=0)
Tdecimate()

But that can be risky in cases where there might be cadence breaks. Or you can lower the CThresh to make it less liable to think something's interlaced when it's not.

[Brad]

Apparently I forgot to thank you guys for the suggestions. Sorry about that, and thanks.

While searching for something else entirely, I stumbled across this other thread from 3 years ago. Guess I should have said that this is from U-matic. I wonder what the cause is.