IVTC from blurred fields

[Wizal]

Sorry for the stupid question but is it possible to IVTC a 29.970 fps movie which has been (wrongly) deinterlaced instead of IVTC in the first place ?

frame 1 : original frame 1 progressive ok
frame 2 : original frame 2 blent with original frame 1
frame 3 : original frame 2 blent with original frame 3
frame 4 : original frame 3 progressive ok
frame 5 : original frame 4 progressive ok

I'm thinking about something like this :

-substract half intensity of frame 1 from frame 2 to obtain half intensity original frame 2
-substract half intensity of frame 4 from frame 3 to obtain half intensity original frame 2
-add both to fully recover original frame 2

The must would be to compare both before adding them since they should be about the same, thus allowing us to detect changes in the original IVTC stream

Would that work ?
Can the currently existing filters do that ?

[SSJ4_Gojita]

Only one way to find out ;) TRY IT!!!

In theory you can IVTC anything, it's just the results you might not be happy with since it might cut out the wrong frames, but you can always manually ivtc.

[DJ Bobo]

Forget it, you won't get anything good

[Richard Berg]

-substract half intensity of frame 1 from frame 2 to obtain half intensity original frame 2
-substract half intensity of frame 4 from frame 3 to obtain half intensity original frame 2
-add both to fully recover original frame 2

The first problem I see is that the contributions to frame 2 will not always be equal -- heck, they could be very very different if this crossed a splice boundary or had fast motion. The information needed to divide them up has been lost.

[Wizal]

The first problem I see is that the contributions to frame 2 will not always be equal
Right, I see your point. No way to compare both halves to detect pattern changes then, but still I don't see why it wouldn't work by just adding them ? (I'm saying really adding here, not a logical AND). If the blur has been done 50-50 there shouldn't be any loss : the substractions of half intensity frame 1 & 3 should remove entirely all data not belonging to frame 2.

Oh well I guess I have to try anyway... yet another filter to develop !!! :rolleyes:

[DJ Bobo]

@ Wizal
You won't even need to try, it won't work. Through blending all field informations are gone and the spaces between what was interlacing lines is filled, not to mention the video having been resized. So you can forget your project, you're just wasting your time.
And no filter can undo this (and there won't be one, even if anyone is crazy enough to even try to develop one)

[tenebrenz]

I, only the other hand, think it is worth investigating. I do think that it is mathmatically possible to certain degree. I think the results that your method would produce would be fairly acceptable though there will be some error. To see how much error somebody needs to try to do it.

If the video hadn't been resized vertically after de-interlacing you could take the idea further. You could take frame 1 and create a new frame by interpolating out the field which isn't present in the blended frame 2. Then subtract half of this interpolated field from frame 2.
Do a similar thing to frame 4 (interpolate out the other field) and subtract half of this from frame 3.
Then half the height of your new frame 2 and 3, weave them together and double the brightness to recreate the orignal frame 2.

I hope the above makes some kind of sense, thinking about fields and frames and ivtc for too long gives me a headache but if I was a developer I would definately be crazy enough to try.

All the best

[Wizal]

@DJ Bobo
I still don't see any major flaw in the theory, so I WILL try :) I think it's quite interesting to have a final word on this considering the huge number of movies floating around with this specific problem. If you really think it's crazy, please be a little more specific on the mathematical point of view.

@tenebrenz
Your approach is interesting, but I guess I'll stick with mine until I can test it. As you mentionned, resizing would be an issue in your method, so I will try to implement mine as a VD filter just to test (no real decimation involved) to see if it's worth it since I have no experience of pure AVISynth filter programming. It might take a while though, because I don't have much time right now :(

Whatever the results are, I will keep you informed on this thread in a few weeks.