Spatial Bob deinterlacer

[Wilbert]

I saw this plugin (under ISC (BSD-like) license) on avisynth.info:

http://pavelsx.googlepages.com/

Readme:

Spatial bob deinterlacing filter similar to spatial part of yadif. Missing lines are synthesized based only on the existing lines of each field using local search for each pixel.

Syntax and parameters:

SBdeint (clip, int mode, int field_first)

mode:

= 0 : full search (default).

= 1 : restricted search - might produce less errors in complex areas at the cost of worse reconstruction of thin near-horizontal lines.


In case you are wondering what ISC means:

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

[Guest]

That's a great filter! Much fewer jaggies than LeakKernelBob() at the expense of a few minor artifacts. It's so much better on jaggies that I can even get away with adding a little sharpening action after bobbing. I'm going to use this on all my sports slo-mo clips now instead of LeakKernelBob(). Thanks for pointing it out.

[Terranigma]

It was posted here as well by the author. They haven't made a post since then though.

[Guest]

I don't care about scharfi's point because my clips have little static content, and even for the little I have it looked fine.

[Terranigma]

Well I just gave it a go for the first time. It seems to be a tad bit faster than the original bob() in avisynth without the added blurriness and better interpolation. Now If only they could continue to work on it and implement static-area-detection/handling; then this would really be a great filter.

[Guest]

Well, he's released the source code, so I may add it myself if the author doesn't want to.

[DeathAngelBR]

Thought I'd share a test I did to compare bob, dgbob, yadif and sbdeint.
It's the most annoying anime I've worked with EVER, with all kind of crappy frames one could find in a freaking dvd.

The script I used (DON'T need to tell me I shouldn't use it like this, because it's the ONLY way this POS looks ok.)

SetMemoryMax(384)
DGDecode_mpeg2source("betterman19.d2v",cpu=4,info=3)

AssumeTFF()
# bob
bob().AssumeTFF()
# dgbob
DGbob(order=1).AssumeTFF()
# yadif
yadif(mode=1,order=1)
# sbdeint
SBdeint(field_first=0)

tf = Selecteven()
bf = Selectodd()
Merge(tf,bf)
movie = last

movie = tfm(movie,cthresh=2,blockx=4,blocky=4).tdecimate()
movie = ColorMatrix(movie,mode="Rec.709->Rec.601")
last = movie
crop( 8, 68, -4, -68)
Undot()
Lanczos4Resize(640,352)
LimitedSharpenFaster(ss_x=1.5, ss_y=1.5, smode=4, strength=125, overshoot=1, undershoot=1)
FastLineDarkenMOD(thinning=0, strength=20)
fft3dfilter()
Tweak(sat=1.10)

First is frame 2415:

BOB() - Kanjis are blured and the background doesn't look too good


DGBOB() - Kanjis look very nice, very sharp, but the background is crap.


YADIF() - Kanjis and background look very nice, though the kanjis with DGBOB look *slightly* better.


SBDEINT() - Kanjis are blured again.


Next is frame 4250:

BOB() - overall it looks good, but a few pixels are off.


DGBOB() - crap.


YADIF() - almost the same as dgbob.


SBDEINT() - excellent.



SBdeint leaves some flickering in lots of frames, other than that it comes out on top. If only I could fix the blury kanjis...