[FFmpeg-user] De-interlacing and Inverse Telecining Questions

[Carl Eugen Hoyos]

Young Kim <shadowing71 <at> gmail.com> writes:

> I've been reading through the ffmpeg documentation 
> about deinterlacing and inverse telecining, and it's
> a bit confusing. In particular, there seems to be a 
> myriad ways of achieving this (i.e. yadif, fieldmatch,
> and kerndeint). Does anyone happen to know what the 
> difference is among these

You first have to understand that while interlaced and 
telecined video look similar they are fundamentally 
different: Deinterlacing means inventing 50% of the 
image where no image was, inverse telecine brings the 
original frames back that were used as input for the 
telecine process.
(Or in other words: The telecine process only duplicates 
some information while interlacing means throwing 
away 50% of the video information - this may even 
happen within camera equipment.)

You should not deinterlace telecined material and you 
cannot inverse telecine interlaced videos.

kerndeint is an old deinterlacer that you should only 
use if you have a specific reason. (performance on arm?)

yadif is a good deinterlacer and has simd optimization 
on x86.

fieldmatch is a modern inverse telecine filter that is 
said to beat MPlayer's pullup filter for every kind of 
source file (but needs more resources).

> and if there is a way to automatically deinterlace / 
> inverse telecine an input accordingly?

Not really because while it is technically possible to 
detect if your video is interlaced or telecined (see 
the idet filter), it is significantly easier and less 
error-prone if you do it (visually) before transcoding.

Carl Eugen