Frameblending in VHS anime

[creaothceann]

I'd like to remove the frameblending from a VHS capture (I don't have access to the VHS). The video is an OVA (Japanese direct-to-video anime) that was released in Italy (i.e. PAL territory).

I get good results with "srestore(frate=24.0/1.001)", but there is still the issue where a bit of the brightness of one frame seems to seep into another frame, like ghosting.

Is there a way to automatically remove that? I don't want to replace these frames manually (via DeleteFrame, DuplicateFrame) unless necessary...

sample 1, 20.35 MB (http://www.mediafire.com/?pz2wcvd1uy7fotf) (from the source file via VirtualDub's "direct stream copy")
sample 2, 12.34 MB (http://www.mediafire.com/?cfvtxp29bpq0mxy) (encoded with ffdshow's HuffYUV)

http://img801.imageshack.us/img801/5246/brightnessdifferences.th.gif (http://img801.imageshack.us/img801/5246/brightnessdifferences.gif) http://img18.imageshack.us/img18/3329/gspotkirara.th.png (http://img18.imageshack.us/img18/3329/gspotkirara.png)

[Usedocne]

I don't think srestore is capable of returning completely artifact free frames, not from what I've seen anyway. The few times that I have used it, have all resulted in some kind of strange "residual luma ghosting".

There probably are ways of getting around it though, but I wouldn't have a clue as to how.

Sorry I couldn't be of more use. :o

[Didée]

I don't think srestore is capable of returning completely artifact free frames, not from what I've seen anyway. The few times that I have used it, have all resulted in some kind of strange "residual luma ghosting".
Srestore can deliver perfectly artifact free frames ... if the source allows. Broken down to the bare bones, Srestore is just a special-purpose decimator: tell the good frames from the bad frames, kick out the bad, keep the good. Hence, it can only spit out what is in the source. Quite often, the "clean" fields are not really clean, but already contain artifacts from the start: residual ghosting from temporal noisereduction (post-normconversion), funky compression artifacts because the compressor had a hard fight compressing the fieldblended interlaced source (and lost the fight), or even ...

... "inverse ghosting", like seen in the sample posted here. I've seen this artifact before - not completely sure about the cause, but possibly it is caused because the interlaced fieldblended frames have been (vertically) sharpened. When a clean field is weaved with a blended field, and the whole thing then is sharpened, the result will be a negative ghost of the blended field is urged into the clean field.

Alternative theory: perhaps it's a "clever" option of the normconversion box, creating this artifact on purpose. (Trying to hide the actual blending by a sort-of "temporal killout".)

Well, this, that, or something completely different. In any case, I don't find a way to repair this. I once had made a more-or-less working filterchain, but that was for an interlaced fieldblended stream. This one here already has been deinterlaced, and because of that, the old approach cannot be used here. Too little information left over. I gave a from-scratch try on the sample, but all ways that actually removed the inverse ghosting also caused destruction in some of the good frames. (Especially those that are animated at full framerate.)


Bottomline: Have much fun with FreezeFrame-ing, and/or YATTA-ing. :)

[creaothceann]

I was afraid of that. :rolleyes: Well, I'm going to use these functions:

DSS2(...)
HQDering.QTGMC(InputType=2).YAHR.srestore(frate=24.0/1.001)
# maybe DFMDeRainbow, too
DSS2(...)
# ...


function n(clip c, int i) {c.DeleteFrame(i).DuplicateFrame(i )} # use next frame
function p(clip c, int i) {c.DeleteFrame(i).DuplicateFrame(i - 1)} # use previous frame
function s(clip c, int i) {c.p(i).n(i + 1)} # scene change

I'll read up on YATTA, too. :thanks:

Btw. here's more info about the anime:
http://www.youtube.com/watch?v=-mQWg3Jj3zU
http://www.mangaupdates.com/series.html?id=915