Deinterlacing. Need help with theory.

[Mystery Keeper]

I'm experimenting with deinterlacing and need help on theoretical side.

I take a frame and throw away all the odd lines. I can restore the lines which have both neighbours, interpolating them. But the bottom line has only one neighbout line and can not be processed that way.

Let's say our half frame has a height of N. I get a picture with a height of 2N-1, but I need 2N. I could just use another interpolation method to resize it to 2N, but that would shift the pre-existed lines, and I really don't want that.

The last line has to be extrapolated using only the info from interpolated to 2N-1 frame. How can I do that?

[AVIL]

You can extrapolate last line from two or three previous lines. There are several matematical algorithms to do this.

E.g. If the value of the pixels located in same column but in n-1 and n-3 are 2 and 4, a value for the pixel extrapolated could be 1

Line->value

N-3 -> 4
N-1 -> 2
N -> 1

Anyway you can deinterlace using neighbour pixels from same frame and temporal contiguous frames. So you have much more information and obtain better results.

Good luck

[Mystery Keeper]

There are several matematical algorithms to do this.

That is what I'm asking for. I need the algorithms or references.

[AVIL]

You can start with the help of wikipedia:

http://en.wikipedia.org/wiki/Extrapolation

http://en.wikipedia.org/wiki/Interpolation

You can also read this thread:

http://forum.doom9.org/showthread.php?t=84725&highlight=mocomped

If your have many interest you can read also this:

www.eurasip.org/Proceedings/Eusipco/Eusipco2005/defevent/papers/cr1859.pdf

http://prestospace.org/training/images/iccvg2004.pdf