Interlaced footage not encoded in interlaced mode: other fixes than Blur(0,1)?

[Chainmax]

Some PAL footage that was encoded to DivX without having been deinterlaced just got to my hands. The only way I could remove the combing artifacts was to use Blur(0,1), but now there's ghosting all over the place. Is there a better solution?

[Mug Funky]

is this resized interlaced or full height?

because if it's full height you should be able to get decent results with a plain ol deinterlacer, but if it's been resized then you're SOL and will have to use blurring (even then there'll still be visible uglies).

[Chainmax]

Well, it's 640x480 @ 25fps so it either has been resized or some funky standards conversion took place. A deinterlacer does no good as you can see in this comparison:

http://img406.imageshack.us/img406/3628/tdeintbefaftnz4.png (http://imageshack.us)

So it seems like I'll be using this then:

Crop(4,2,632,476,align=true)

Spline36Resize(656,448)

Blur(0,1)

FFT3DFilter(sigma=3,plane=3,bw=32,bh=32,bt=3,ow=16,oh=16)

DeBlock_QED(quant1=35,aOff1=16,quant2=45,aOff2=6)

DeGrainMedian()

gradfun2db()

Tweak(sat=1.2,bright=-10)

AddGrain(5,0,0)

AddBorders(32,16,32,16)

Levels(0,1,255,16,235)

ConvertToYUY2()

Scene changes and parts with heavy combing look awfully blended as a result of the blurring, could something like CDeBlend or MRestore be used to alleviate that?

[Mug Funky]

aah. it's been resized (ever so slightly).

2 options:

1. spend a lot of time trying to figure out the resize factor from clues in the source, then resize back to the original size and deinterlace. use a resizer that best mimics the original one used (again, trial and error). you may get away with some blurring, but a lot of intact motion (imagine doing a blend deinterlace, then mixing it at ~30% opacity with the original, then bobbing). you may also get nothing useful.

2. use a blend deinterlace, or maybe vinverse.

option 1 is lots of work and may yield nearly identical results to option 2...

[Chainmax]

I thought about using Vinverse but couldn't find a download link in this forum. Does it have a site where I can be sure to bownload the latest version from?

[henryho_hk]

Such blocks are very hard to deal with. Try experiments with the interlaced mode of blindpp() of dgdecode.dll or even fft3dfilter() followed by a deinterlacer. Smooth deinterlacers such as tomsmocmp() w/ vertical filter may work better.
----
Oops, I did not know that it's an interlaced source resized progressively..... Thanks Mug for the tips.

[Boulder]

I thought about using Vinverse but couldn't find a download link in this forum. Does it have a site where I can be sure to bownload the latest version from?http://bengal.missouri.edu/~kes25c/vinverse.zip

Looks like tritical never added it on his page, I found the link in the TDeint/TIVTC thread.

[foxyshadis]

I bet Momonster could create a script that could unblend that into halfway decent 60fps. =p I know he has a lot of other things to work on though.

[wonkey_monkey]

Could you post a short clip? I'm interested to see what can be recovered, although it might be too far gone after the encoding.

David

[scharfis_brain]

just blur that clip. everything else is just hopeless.
I tried that several times before without success.

after blurring you may appy limitedsharpen to make it look a little bit more crispy.

[Chainmax]

Vinverse didn't do much good, I'd put its effect somewhere in between the picture I posted (in which TDeint() was used) and using Blur(0,1). Switching to TDeint(type=4), however, seems to yield the exact same effect as the script I posted: it doesn't even seem sharper in spite of ditching the Blur line, but that probably is just my imagination. So, the script is now:

TDeint(type=4)

DeBlock_QED(quant1=35,aOff1=16,quant2=45,aOff2=6)

Crop(4,2,632,476,align=true)

FFT3DFilter(sigma=3,plane=3,bw=32,bh=32,bt=3,ow=16,oh=16)

DeGrainMedian()

Spline36Resize(656,448)

dull=last
sharp=dull.LimitedSharpenFaster(SMode=4,Strength=150)
Soothe(sharp,dull,25)

gradfun2db()

Tweak(sat=1.2,bright=-10)

AddGrain(5,0,0)

AddBorders(32,16,32,16)

Levels(0,1,255,16,235)

ConvertToYUY2()

scharfis_brain: yeah, blurring and/or blend deinterlacing seem the only way to make it watchable. I had previously ditched LSF because it made some extremely blocky scenes quite bad and whatever extra sharpness it brought was buried by all the blurring and smoothing that came before. With this new filterchain, however, it actually makes a nice difference that is worth a few bad scenes :).

davidhorman: probably not, the footage is of a sensitive nature. We can further discuss it via PM if you want.

Everyone: there is no way to fight back the ensuing ghosting, is there?

[wonkey_monkey]

davidhorman: probably not, the footage is of a sensitive nature. We can further discuss it via PM if you want.

That's okay, I'm more curious than hopeful ;) I found that resizing down to 80% (from 270 to 216 with the sample, from 480 to 384 with your source) and there was just enough separation to see two fields, but I don't know how this frame compares to the rest of the source.

David