Possible to use Masktools to enhance GuavaComb?

[tedkunich]

I'm working on a LD capture project and have hit a bit of a snag. I get quite a bit of chroma cross talk into the luma creating the infamous NTSC checkerboard pattern in stationary edges and bright red or blue objects. I have tried virtually all combinations of settings in GuavaComb without much success - I can reduce/eliminate the hatching at the expense of introducing artifacts whenever there is any motion elsewere in the frame but adjusting the setting to eliminate the artifacts render the filter useless.

Would it be possible to use something like an edge mask or a motion mask to apply GuavaComb only to truely stationary scenes yet keep it from introducing artifacts in areas with motion? I downloaded the masktools 1.5.1 package but have no experience with it - the parameters in the examples at the bottom of the documentation do not match with the actual function documentation. I played a little with the edgemask and motion mask filters but my output alternated from a black and white mask to a green tinted one, to one with rainbows as I stepped through the frames.... It did not seem correct... Can anyone more experienced with these tools give some guidance to a noob??? :)

FWIW, my playtime with the masktools was on interlaced, non-IVTC video, because I need to perform the Guava filtering before my IVTC step.


Thanks

T

[oo_void]

I too have found that GuavaComb just doesn't seem to cut it for me. (maybe I'm just using it wrong though). I'm finding that I have to turn it up so high to get the desired effect, that I start getting unwanted artifacts in other regions of the image... usually sploches of grey in areas of blurred, fast motion. Maybe it doesn't like BFF sources...

I've also toyed around with importing the VirtualDub Dot Crawl filter, but it too has problems, especially with scene changes. I can turn of the Temporal Filtering, but then Dot Crawl just seems to blur things too much.

I think I may have found something interesting though, it's called DeDot() and can be found here.. http://nullinfo.s21.xrea.com/. Unfortunately, all the information about the filter is in Japanese (anyone have some insight on the params it takes?), but in playing with it last night, I'm was getting great results with even the default settings.

I was just going to run it through a SeeTheDifference type function to find out what's really going on, but I can't seem to RDP into my rendering box today... I guess it'll have to wait for tonight. DeDot() definately worth looking into as an alternative to GuavaComb though.

------
oo_void

[JuanC]

Originally posted by tedkunich
I'm working on a LD capture project ... Have you tried capturing the video using the S-Video connector? It should help a lot avoiding luma-chroma interference.

Also, after reading this post from High Speed Dubb (http://forum.doom9.org/showthread.php?s=&postid=365047#post365047) (and the whole thread) I am always using guavacomb and peachsmoother for my analog TV captures as the first filters in my AVS scripts, something like this: <whatever>Source( ... regularly YV12 ...)
crop( ... your settings ... )
ConvertToYUY2(interlaced=true)
SeparateFields()
GuavaComb( ... your settings ... )
PeachSmoother ( ... your settings ... )
Weave()
... Your other filters ... And I'm getting good results even with the default parameters. No visible artifacts in motion scenes, some rainbows are gone and dot crawl is completely gone in slow scenes!

[tedkunich]

@oo_void

DUDE!!!! great find!!!!

This thing works great!!!! Now I can start the long filter/render session instead of mucking with filters!!!

Thanks,

T

[JuanC]

Originally posted by tedkunich
DUDE!!!! great find!!!!

This thing works great!!!! Now I can start the long filter/render session instead of mucking with filters!!!

Thanks,

T :D Which one? The DeDot filter, the S-video connector or the separatefields - weave thing?

[tedkunich]

I am capturing on the SVid port - my LD player does not have all that great of a comb filter in it. I was almost going to recapture everthing with the composite-in as I can play with parameters in my capture card, but it is not as sharp of picture as the Svid gives - some low pass filtering goin on there...

Tried the separate fields trick, but as I suspected, it would not take care of this... once you break up the interleaved NTSC pattern, either via separatefields or IVTC, the algorithms in Guava go to crap.... If you are getting acceptable results, I would argue then that your other filtering is removing any dot crawl, not Guava.

The Dedot filter that oo_void posted works like a charm - the filter kicks in within a few frames and does not produce any of the speckles that Guava produced in motion scenes.... Try it out... works fantastice with just the default parameters.... (or course I'll play with 'em to eek out that last bit.... ;)

T




Originally posted by JuanC
Have you tried capturing the video using the S-Video connector? It should help a lot avoiding luma-chroma interference.

Also, after reading this post from High Speed Dubb (http://forum.doom9.org/showthread.php?s=&postid=365047#post365047) (and the whole thread) I am always using guavacomb and peachsmoother for my analog TV captures as the first filters in my AVS scripts, something like this: <whatever>Source( ... regularly YV12 ...)
crop( ... your settings ... )
ConvertToYUY2(interlaced=true)
SeparateFields()
GuavaComb( ... your settings ... )
PeachSmoother ( ... your settings ... )
Weave()
... Your other filters ... And I'm getting good results even with the default parameters. No visible artifacts in motion scenes, some rainbows are gone and dot crawl is completely gone in slow scenes!

[tedkunich]

Originally posted by JuanC
:D Which one? The DeDot filter, the S-video connector or the separatefields - weave thing?

Missed by that >< much!

See above

[tedkunich]

[Edit] Damn Geocities... can't direct link to a photo...

Hope that this works....

Here is an example of the NTSC crawl that I'm trying to get rid of:

Look closely at the red "post" in the background - it has this annoying checkerboard pattern... it has always bugged me watching the LD on my TV! I have not performed any filtering other than the dedot filter in either of the clips.

Photo compare (http://www.geocities.com/tedk_1/dedot.htm)

I'm impressed.... Or friends in Japan have been holding out on us!!! ;)



T

[oo_void]

Glad I could be of help to someone out there...

For JuanC, my capture was done using SVHS into my DV cam, but I was still getting some nasty crawl at times. To be fair to GuavaComb though, the scenes that were giving it the most trouble were really nasty. Bright colors on black that also exhibit a bit of shaking (Star Wars was done before the digital days). Here are a couple samples stills ...

DeDot Sample (http://www.badassgeek.com/img/dedot_sample.png)
GuavaComb Sample (http://www.badassgeek.com/img/guava_sample.png)

For the GuavaComb sample, I had to let the video run a bit before any difference appeared (that's what a temporal filters do, right ;)), the DeDot Sample is just a single frame with no pre-roll. You'd really have to see things in motion for an effective comparison though.

------
oo_void

[scrtagt69]

oo_void, just wondering man, did ever fine what params to use with dedot, i used it w/o any and it works waaaaay better than guava. and for my case, i also used DeGrainMedian(limitY=5,limitUV=10,mode=0) and it worked very well for me. good find.

[LB]

I've been using dedot for about a year and a half now. Little known amazing filter. Try these settings: DeDot(20,20,15,5) and the trick is to ALWAYS ALWAYS ALWAYS use dedot BEFORE BEFORE BEFORE doing any ivtc or anything else to the video. That's crucial.

[oo_void]

I'll second that, always run your comb filter before screwing with the video signal. But, I have a trick question... Does anyone have any insight as to whether or not this plugin has a field order preference (or comb filters in general)?

Most of my caps are DV based, that means BFF... Should I switch the field dominance to TFF before running a comb filter (given that the filter might be expecting the more standard field order), leave it as is, or does it not matter for comb filters?

------
oo_void