Deinterlace test - no winner?

[agilpwc]

I have a camera that shoots 1440x1080 60i and most of the stuff I shoot has a lot of motion. I wanted to find out the best deinterlacer for my video. Which in the end I convert to 1280x720 30p.

I did a comparison of leakkerneldeint, yadif, YV12InterlacedSelectTopFields (which justs discards a field).

I have two frames. In the first frame the camera was moving, in the second the camera was still. I thought I would see more detail with the deinterlacers over discarding a field in the still shot. But wasn't really the case. With my untrained eye I could not tell any difference between yadif and leakkernel.

I zipped up the frame captures if any one wants to look and give their "professional" opinion.
http://avs-scratch.s3.amazonaws.com/deinterlace-test.zip

[foxyshadis]

No comparison of deinterlacers is possible without a video segment, a single-frame capture is fairly worthless for testing.

For top quality deinterlacing where speed is no issue, look at MCBob (or the older MVBob). You're only using very fast ones, many more exist, TDeint is a medium between yadif and the MVs.

[scharfis_brain]

you got lesser noise and slightly more detail in static shots.

if you need a quick conversion with fairly good results, I suggest:

last.yadifmod(edeint=last.bob().selecteven())

I don't recommend leakkerneldeint, cause it has a weak motion mask.
See leakkerneldeint-272.png the windows of the car,
or leakkerneldeint.png the legs of the men in the middle of the screen.

I also don't recommend to completely discard one field, cause you'll end up - as I already mentioned - with more noise and lesser fine detail (better defined diagonals).

[agilpwc]

Thanks for the suggestions. Speed is an issue for me, and slow deint are out of the question.

I don't understand how watching a video is better than still frames for comparison? I can hardly see a difference looking at still frames, I can't imagine trying to compare to multiple videos.

I see what you mean about leakkerneldeint.

I don't understand the comment about not discarding fields. I see that if you wanted to resize to 1920x1080 then discarding fields would be stupid. But if your final output was small enough like 960x540 or smaller than discarding field would be the way to go right?

[scharfis_brain]

it would even be better to properly deinterlace, cause you'll reduce aliasing and stairstepping.

[agilpwc]

That makes sense about not discarding field, thanks scharfis_brain.

It is amazing how there is to learn about video if one is so inclined.

[Terka]

try mvbobmod

[Roco]

I have a camera that shoots 1440x1080 60i and most of the stuff I shoot has a lot of motion. I wanted to find out the best deinterlacer for my video. Which in the end I convert to 1280x720 30p.

Why do you want to reduce the frame rate?
I also have such a camera (Z1) but instead of reducing the frame rate, I prefer to convert fields to frames, so I end up with 60p.
This way, I don't loose information, since information of each field is 100% present on the screen. Only "disadvantage" is some slight flickering in rare cases but there is the huge advantage of the smoothest motion possible. Also, this process is fast too.

This is a 60p full HD sample (converted). Set your monitor at 60fps to watch it at 100% smooth motion. Preferred player the "Media Player Classic" with settings Options/Overlay, Options/Internal filters/All left panel filters checked and All right panel filters checked, except mpeg-1.
http://rapidshare.com/files/86122777/Summer_1080p60_50Mb_ac3128kb.mpg

Recompression to 1080p60 mpeg2 discussion:
http://forum.doom9.org/showthread.php?t=133687&highlight=1080p60

Alternatively, I would recommend 720p60 (1280x720 @ 60fps).

To administrator:
There is something strange with this thread (on the above link). I can only find it through the search engine!