Field-based ghosting from camcorder causing problems when bobbed

[ShawnFumo]

I've just encountered a weird problem with the new camcorder I just bought. I don't know if it is doing something weird with its internal gain or something, but I may have to bring it back if I can't fix this.

I've been filming some instructional videos using a friend's camcorder previously (an old Sony Digital-8 model) and I found the best solution was LeakKernalDeint for normal motion and LeakKernalBob with assumefps to do smooth slow motion. Both with threashold 0 due to the nature of the video. It worked well since I wanted to give people the option to look at the video frame by frame and LeakKernalBob gave extra information. It was pretty noisy indoors, but FFT3DFilter helped a lot on that end.

So now I just bought a Panasonic PV-GS31 and the features and quality seem pretty good (lots of manual control).The output seems to look decent interlaced or during a normal deinterlace.

The problem is that as soon as I use any kind of bob deinterlace (leakkernalbob, bob, or whatever), it has a pretty severe ghosting effect. On very fast motion, sometimes I see three or more faint after-images trailing after the object.

Here are two short clips showing the output. It isn't quite as obvious once compressed to xvid, but you should still be able to see it if you look carefully or go frame by frame at the end of the bobbed video:

Normal Deinterlaced at 15fps (http://video.yoyoing.com:8080/yoyoing/shawn/videofiles/camcorder/deintoutxvid1.avi)

Bob Deinterlaced at 30fps (http://video.yoyoing.com:8080/yoyoing/shawn/videofiles/camcorder/boboutxvid1.avi)

Any ideas on if there's a way to correct this in post? It'd probably be solved by much more light, but it is convenient the way I'm doing it now and I never had this issue with the old camcorder. It is pretty annoying..

Thanks!
Shawn

[scharfis_brain]

please provide the original, unprocessed video and the scripts you used for 15 fps and 30 fps clip.

[ShawnFumo]

Thanks Scharfis,

Here is a slightly longer clip of the original DV:
clip2.avi (http://video.yoyoing.com:8080/yoyoing/shawn/videofiles/camcorder/clip2.avi)

The script for normal deinterlace:

bigclip = avisource("clip2.avi").ConvertToYV12(interlaced=true)
bigclip = LeakKernelDeint(bigclip, 0, threshold=0)
bigclip = assumefps(bigclip, 15)
return bigclip


and bobbed:

bigclip = avisource("clip2.avi").ConvertToYV12(interlaced=true)
bigclip = LeakKernelBob(bigclip, 0, threshold=0)
bigclip = assumefps(bigclip, 29.97)
return bigclip


Of course normally I'd have the regular deinterlace going at 29.97 fps also, but just did that here to make it easier to compare the two. It seems like there's a very slight ghosting even in the normal deinterlace, but that the bobbing makes it more severe.

Shawn

[ShawnFumo]

Also, here is another clip I filmed this morning which is brighter but still exhibits the issue when bobbed. Also, the noise itself seems to strobe after the bob a bit:quatro_clip1.avi (http://video.yoyoing.com:8080/yoyoing/shawn/videofiles/camcorder/quatre_clip1.avi)

Thanks,
Shawn

[ShawnFumo]

Ok, I figured out a bit more of what is going on. I can also now see why the bobbing is causing the trailing to be so obvious.

Check out this picture (You may want to save these to your hard drive because the effect seems diminished in IE with the bright background and this way you could flip through them with something like the default Windows viewer). It is straight from a simple separateFields. You can see two after-images above the string like spokes of a bicycle wheel (I was doing a circular motion).

spokes1.png:
http://www.yoyoing.com/shawn/videofiles/camcorder/spokes1.png

Now the following picture is from the same field of the next interlaced frame. Notice how the "spokes" are in about the same place but have faded. This is why it isn't that noticable in a normal deinterlace.

spokes3.png:
http://www.yoyoing.com/shawn/videofiles/camcorder/spokes3.png

Lastly, look at this picture. This is from the other field of the first interlaced frame (between the last two time-wise). Notice how these spokes are in between the others. This is why it is so noticable when bobbed. Instead of staying in one place, the afterimages strobe, which attracts your eye to it.

spokes2.png:
http://www.yoyoing.com/shawn/videofiles/camcorder/spokes1.png

If you look at the images in sequential order (1,2,3 instead of how I have it in this post), you'll see that the newest spoke in #3 is where the string was in #1. But it isn't there at all in #2! So it isn't from that moment in time like real motion blur, but is from the last time that field was imaged. Is a CCD imaged entirely at 60fps with half the fields thrown out each time? Or do the sensors work at 30fps and alternate? If the former, I can only think the GS31 has some sort of "smart" electronic gain feature which isn't that noticable when interlaced. If the latter, could it be the CCD sensors take a sec to power down from pure white?

Has anyone encountered this before???
Shawn