Query about identifying field-blending

[pooksahib]

Hi. I'm gradually teaching myself about restoration with the help of you good people. After going over what I've learned so far, I have a question about the one topic that I find most difficult – identifying what needs to be done to a DVD. Attached is a page from my own manual. With the clip opened in Vdub and 'bob doubler' activated, there are two identical frames for each image and the combing is plain to see.

What I'd like to know is, can the fact that it was field-blended be confirmed (or ruled out) at this stage? Or is it necessary to check the effects of QTGMC before deciding? Thank you.

[jagabo]

With the clip opened in Vdub and 'bob doubler' activated, there are two identical frames for each image and the combing is plain to see.

If both fields are identical (aside from comb artifacts from the crude deinterlacing), and both show the same blending, that would be frame blended, not field blended.

[pooksahib]

OK, thanks. The clip is in post #4 of this thread:
http://forum.videohelp.com/threads/377869-MeGUI-my-first-use-of-it-and-some-questions?...05#post2439605
and in post #7, manono tells me that "it was converted to PAL from a film source by way of field-blending".

Can you confirm frame- or field-blending (if you're willing)?

If it is 'frame', is SRestore still the answer? Or another filter?

[jagabo]

That's field blending but it's different than the usual NTSC/PAL conversion field blend. SRestore() won't work well on it. I don't know of any filter that will work well with that type of field blending.

[pooksahib]

Thanks again.

Here's something odd I've just discovered. As I said above, with the clip opened in VDub and 'bob doubler' activated, there are two identical frames for each image. But when opened in MeGUI with 'Yadif (with bob)' as the filter, every single frame (and there's twice the original number) is different.

Aren't 'bob doubler' and 'yadif (with bob)' effectively the same thing? Is the fact that one shows pairs of identical frames while the other shows unique frames mean I'm getting conflicting information about the nature of the source material?

[jagabo]

with the clip opened in VDub and 'bob doubler' activated, there are two identical frames for each image.

No, there are not. The fields alternate between a clear and blended. Are you looking at the correct pane in VirtualDub? The output pane shows the bob doubled image.

[pooksahib]

Damn. Just discovered the input pane/output pane/both panes options. Colour me red-faced. Thanks.

As regards SRestore, I'll probably use it here as it seems to have done more good than harm (see the triptych above).

[Anonymous344]

After calling SeparateFields(), all the even fields look to be blends of the (clean) fields on either side. If so, why not discard the even fields and resize the odd ones? You're not losing anything.

MPEG2Source()
SeparateFields()
SelectOdd()
Spline36Resize(720,576)
AssumeFrameBased()

The assumption is that the whole video is like the sample and that there's no change in the pattern.

[jagabo]

After calling SeparateFields(), all the even fields look to be blends of the (clean) fields on either side.

But that's not the case. It alternates with the even fields blended for a while, then the odd fields, then back to the even fields, etc. This type of field blending is sometimes used for 24p to 50i.

[Anonymous344]

But that's not the case. It alternates with the even fields blended for a while, then the odd fields, then back to the even fields, etc.

In that case, I overlooked the transition. It's a shame that it alternates. Perhaps there's some way of auto-detecting the blended fields and removing them, leaving the clean ones, which could then be resized.

This type of field blending is sometimes used for 24p to 50i.

I don't see the benefit in doing that. The result looks awful.

[jagabo]

Perhaps there's some way of auto-detecting the blended fields and removing them, leaving the clean ones

Not really. Because halfway between the point where one field is clear, and the point where the other field is clear, both fields are equally blended. So in the those areas there are no clean fields to chose from. Here's a sample script that first builds a black image with white vertical bars, animates the bars so they are moving to the right, then performs the type of frame rate conversion used in the sample clip:

# build a black image with white vertical bars
BlankClip(width=1440, height=480).ConvertToYUY2()
whitebar=Crop(0,0,16, height).ColorYUV(off_y=220)
Overlay(last,whitebar, 0, 0)
Overlay(last,whitebar,  360, 0)
Overlay(last,whitebar,  720, 0)
Overlay(last,whitebar, 1080, 0)

# animate the image so the bars move to the right, 24 pixels per frame
Animate(0, 30, "Crop", 720,0,720,height,true,  0,0,720,height,true)
Trim(0,29)
last+last
last+last

#apply tff and 24 fps properties
AssumeTFF()
AssumeFPS(24)

# convert to 50 fields per second with field blending
ConvertFPS(50)
ShowFrameNumber(x=10, y=30)
SeparateFields()
SelectEvery(4,0,3)
Weave()
# at this point we have a 25i video with field blending like the sample video

# for clarity stack the two fields vertically instead of weaving them
SeparateFields()
StackVertical(SelectEven(), SelectOdd())

You can see that the top field starts out clean at frame 0, and the bottom field is blended:



By frame 12 the top field is blended and the bottom field is clean:



But half way in between, around frame 6 (fields 12 and 13), both fields are blended:



In those regions there are no clean frames to choose from so the usual techniques using SRestore() will leave a lot of those lesser blended frames.

[pooksahib]

Interesting stuff and thanks for your time. The good news is that the resulting vid after using QTGMC and SRestore is perfectly watchable so I won't be doing any more work on it.

[Anonymous344]

Not really. Because halfway between the point where one field is clear, and the point where the other field is clear, both fields are equally blended. So in the those areas there are no clean fields to chose from.

Oh, I see. Judging by the result of your sample script, it's not just a single frame at the halfway point either: the fields before and after 12 and 13 look blended to a lesser degree. What a mess!

[jagabo]

it's not just a single frame at the halfway point either: the fields before and after 12 and 13 look blended to a lesser degree.

Yes, over a period of 25 frames the top field goes from clear to slightly blended, to most blended, to slightly blended, then back to clear. The bottom field does the same but 90 degrees out of phase -- when the top field is clear the bottom field is most blended, and vice versa.

[Anonymous344]

Thanks! The fact that the transition from clear to most blended is gradual led me to misinterpret what I was seeing. I appreciate your explanation, jagabo. I've never seen a video quite like this one before, and it's always good to learn something new.

[bruno321]

After calling SeparateFields(), all the even fields look to be blends of the (clean) fields on either side. If so, why not discard the even fields and resize the odd ones? You're not losing anything.

MPEG2Source()
SeparateFields()
SelectOdd()
Spline36Resize(720,576)
AssumeFrameBased()

The assumption is that the whole video is like the sample and that there's no change in the pattern.

Just cueing in to say this was useful for the case I'm at. I have an NTSC source with field blends which srestore doesn't take care of correctly. The pattern is: BGBGR, where B is a blended field, G is a good field, and R is a repeated field. What worked was a variation of the above:

SeparateFields()
SelectEvery(5,2,4)
Spline36Resize(720,480)
AssumeFrameBased()

Just posting it cause it may be useful to someone in the future.

Cheers,