Are the fields of an interlaced really "shifted"?
 

Straightforward question:

In an interlaced video, if i swap the fields and I invert the field order is the result the same?

I reformulate the question in a different way.
Do the even lines and the odd lines represent different lines "in space" other than in time?
If the analogue cameras do record a field "seeing" the same image when they record the other field the answer should be yes. Otherwise if the analogue cameras see the second field "shifted" by 1 line the answer should be no.

I'm afraid to be not clear enough so I ask the question in another equivalent way:
If i separate the fields and I interleave them, obtaining a double frame rate video with half the height, should the odd frames be a half pixel shifted down (even if it is impossible) to make the video really stable?

Does the odd lines intrinsecally contain information relative to a lower part of the image in respect of the corresponding even lines?

Thanks!

That's not exactly straightforward, LOL! Glad you rephrased it below. But if you invert (reverse) the field order, motion will be back-and-forth instead of smooth. An object moving left to right will move ahead-back-ahead-back instead of ahead-ahead-ahead-ahead.

Top and bottom fields occupy different scanlines. Scanline numbers begin at line 0. In top field first video (TFF), the top field plays first and occupies scanlines 0,2,4,6,8..., etc. The bottom field plays arfterward and occupies lines 1,3,5,7...etc.
In bottom field first video (BFF, usually DV format), the bottom field plays first and occupies lines 1,3,5,7,9..etc. The top BFF field is the second to play and occupies lines 0,2,4,6,8...etc.

Neither. The camera doesn't "see" shifted fields, and doesn't see the same image at two points in time, even if nothing moves.The camera sees two images at two different points in time. The circuitry doesn't shift the fields, it records the two images on two different sets of scanlines in the same frame.

BTW, digital cameras that record interlaced video work the same way. You've heard of the DV format? The same is true of newer digital cameras that record h.264 interlaced video.

If you separate the fields they are already interleaved, one after the other. You will double the frame rate and the images will be half-height. The alternating fields will appear to shift upward one pixel when playing. The image will bob up and down during play and will not be stable.

When fields are separated as in SeparateFields, the top and bottom fields contain only the information in their respective images. They don't contain information that is missing in part of another image, if that's what you mean. If you use SeparateFields() on a video that is 480 pixels high, each half-height field will contain 240 lines. If you use SeparateFields() on a video that is 576 pixels high, each half-height field will contain 288 lines.


If you want to stabilize a video that has up/down displacement, deinterlace to full-size frames , run a stablizer on it and reinterlace. If you have a defective video that has uneven vertical spacing between fields, you'll need Avisynth to fix it (if possible).

Thank you a lot for the very clear answer!

Yes this is the former reason why I posed the question!
I suspected that the only way to stabilize an interlaced video was to deinterlace->stabilize-->reinterlace, but... Doesn't this procedure make the final video full of "interpolated junk" that it's not the real original information (but shifted in order to be stabilized)?

If you can detect distortion in a video that is stabilized by a mild filter like Stab(), you're really good. Even a stronger stabilizer like VirtualDub's DeShaker does very clean work. Something stronger like drastic settings in other stabilizer plugins will likely produced some form of distortion, but the principle technique used is some form of resizing. You will see artifacts in any motion-compensated interpreter, even with QTGMC. The choice is between some imperfection or watching frames hop up and down.