Black Edge In Dv Movie

[user_deleted]

Hi,
I'm trying to key a DV image with Primatte, and no matter what, there is always a black edge on the left side of the foreground object. Is there any way to get rid of it? I tried to colorcorrect it, but it would always affect the whole object, not only the edge itself. I couldn't find any topics dealing with this problem.

[user_deleted]

that's a very common DV keying problem...

make sure you reduce the compression artifacts by chroma smoothing or with the 4:x:xEnhancer macro, and use the alpha mask to color correct only the edges if needed (shrink, invert, blur, or use the alphaEdge macro). lightwrap to finish.

++ christoph ++

[deke-kincaid]

Just do a Colorspace(RGB > YUV), Blur the B channel 5-10, then a Colorspace(YUV > RGB)

Also there is a DVBlur Macro that essentially does this.

[user_deleted]

QUOTE(beaker @ 03/01/06, 06:10 PM) [snapback]229626[/snapback]
Just do a Colorspace(RGB > YUV), Blur the B channel 5-10, then a Colorspace(YUV > RGB)

i'm curious, why not blur the G (ie U in YUV) channel as well?
++ christoph ++

[deke-kincaid]

woops, yea, I meant blur G, not B.

[user_deleted]

well, i usually blur both channnels anyway..
even if i dont have to do any keying
++ christoph ++

[deke-kincaid]

Well U is the B-Y(Blue Yellow) channel. Blue is always the weakest link on most mediums(film, video, scanners, etc..) because the waves are longer and thinner. Technology in the past trys to equally sample RGB which is a bad idea because of the different wavelenghts of the different colors and not all colors are created equal. R ends up getting over sampled(thats why it can blow out easily on television) and B always gets undersampled, G is just right.

Thats why in dv it is better to shoot stuff with a PAL camera since it is 4:2:0 instead of NTSC 4:1:1. THe B-Y channel needs that extra sampling to get the quality up closer to R-Y. Luckily HDV went with 4:2:0.

V (R-Y) channel rarely has issues. Blurring V only softens the picture.

[user_deleted]

hmm.. not sure if you dont mix stuff up here.

the U channel is the B-Y (which has nothing to do with yellow, but is the blue-luminance difference) and V is the R-Y channel (which is the red-liminance difference). green gets reconstructed from the two and the full luminance signal.
also, PALs 4:2:0 sampling doesnt mean that there is more U than V samples, because it alternates (ie 4:2:0 then 4:0:2) it just means that it makes square 2x2 macro blocks intead of NTSC's rectangular 4x1. the blue channels is usually the worst on video because it gets assigned the least luminance, but that's true for 4:2:2 sampling as well.

anyway, in my experience blurring the U channel will smooth mostly the blue channel, blurring the V channel will smooth the reds and the greens. so you could choose to blur only the appropiate channel for bluescreen (U) or greenscreen (V).. but blurring U and V will smooth all color channels which looks much more pleasing while hardly softening the image.

for what it's worth
++ christoph ++

[deke-kincaid]

Yea, B-Y = Blue minus Luminance and R-Y Red minus Luminance. More of me answering questions when I'm tired and retarded

[user_deleted]

In my experience, the black edge round one side of the FG in a comp is an artifact of the various deartifacting techniqes discussed above and not an artifact of the 4:2:0 (or 4:1:1) compression. DV compression results in blocky edges to the matte.

These deartifacting techniques fix the blocky matte problem but create a dark line instead. The lesson to be learnt is that DV is not suitable for visual effects work.

You might want to try this as a dirty fix:

REPLACE_WITH_FOOTAGE = ColorWheel(720, 576, 2, 0, 1, 1, 1);
ADJUST_ME = Brightness(REPLACE_WITH_FOOTAGE, 1.5);
Reorder1 = Reorder(REPLACE_WITH_FOOTAGE, "aaan");
DilateErode3 = DilateErode(Reorder1, "rgb", 1, xPixels/GetDefaultAspect(),
0, 0, 0);
Emboss1 = Emboss(DilateErode3, 10, 180, 0);
Expand1 = Expand(Emboss1, 0, 0, 0, 0, 0.6875, 0.6875, 0.6875,
1);
DilateErode2 = DilateErode(Expand1, "rgba", 2.5, xPixels/GetDefaultAspect(),
0, 1, 0.625);
Blur4 = Blur(DilateErode2, 12, xPixels/GetDefaultAspect(), 0,
"gauss", xFilter, "r");
ISub1 = ISub(Blur4, Reorder1, 1, 100);
KeyMix1 = KeyMix(REPLACE_WITH_FOOTAGE, ADJUST_ME, ISub1, 1, "R",
100, 0);

[user_deleted]

QUOTE(mikemarcus @ 03/05/06, 03:18 PM) [snapback]230010[/snapback]
In my experience, the black edge round one side of the FG in a comp is an artifact of the various deartifacting techniqes discussed above and not an artifact of the 4:2:0 (or 4:1:1) compression.

well, i've seen dark edges even on digibeta footage, or DVCPRO HD (because somebody forgot to to turn down the detail or something).. but dv deartifacting certainly doesn't make things prettier.

but some questions about your fix:
(btw, it's great that you post real scripts, good stuff to learn)

QUOTE
Reorder1 = Reorder(REPLACE_WITH_FOOTAGE, "aaan");

i take it that the final alpha channel from the key should be used as an input here?
(ok, that's a stupid question, but just to be sure)

QUOTE
Emboss1 = Emboss(DilateErode3, 10, 180, 0);

what's the advantage of an emboss here over an edge detect?
or why not an alphaEdge macro in the first place?

QUOTE
Blur4 = Blur(DilateErode2, 12, xPixels/GetDefaultAspect(), 0, "gauss", xFilter, "r");

hmm.. i guess that you only blur the "r" channel to speed up the script (as you use the "r" input on the keymix). but wouldnt it be faster to blur it on "rgb" since that keeps the iSub as a 1 channel image.
ok, that's another stupid question as i can test it myself...
result: seems to take about the same time

++ christoph ++

[deke-kincaid]

QUOTE(mikemarcus @ 03/05/06, 06:18 AM) [snapback]230010[/snapback]
These deartifacting techniques fix the blocky matte problem but create a dark line instead. The lesson to be learnt is that DV is not suitable for visual effects work.
Tell that to the clients Wait till the large proliferation of HDV starts going around. Had a tv show looking to cut the budget by moving to all HDV. Luckily they stuck with HDCAM (though hdcam's 3:1:1 sucks too). There we're some tards "advising" them that moving to HDV would be just fine.

[user_deleted]

Beaker: My opinnion is that a VFX soup should get involved during preproduction. If this happens then they can say "the money you save by shooting on DV (or whatever) is nothing compared to the money that it will cost you in post to fix the problems that you created and even then, the quality of the composites will suffer badly"

Christoph: The reorder is because emboss will only work on rgba channels. It takes the alpha from the key and puts it in RGB. I should have written "replaice with keyed matte" rather than "replace with footage".

the emboss in this case is used a one sided edge detect (the dark edge is usually on one side of the key)

I only care about the red channel because I only use that one. I could just as easily blurred the green channel and used that for the keymix. It might be more efficient to blur rgb, to be honest I never think of images in terms of BWA always RGBA, I dont really know how shake deals with one channel images. Now that in version 4 the noodles change colour to signify what channels are what, I should really get to grips with it.

Note that any color correction that is nessicary can go above the second input of the keymix, the great thing is that they will all concatenate.

Note that I havent tested this script on footage, i just knocked it up as an example. It will almost certainly need some tweaking.

[user_deleted]

QUOTE(mikemarcus @ 03/06/06, 10:44 AM) [snapback]230073[/snapback]
the emboss in this case is used a one sided edge detect (the dark edge is usually on one side of the key)

ah, good idea, although i find that often the black edge only apears only on parts of the foreground object (and unfortunately not always on the same side).. so i sometimes try a masked alphaEdge.

QUOTE
I only care about the red channel because I only use that one. I could just as easily blurred the green channel and used that for the keymix. It might be more efficient to blur rgb, to be honest I never think of images in terms of BWA always RGBA, I dont really know how shake deals with one channel images.

from my understanding 1 channel (BW) images process 3 times faster than 3 channel (RGB). however, since shake seems to be clever enough to see to decide internally which channel is really used in the end, it often doesn't seem to make a difference as only the necessary channel gets processed in the first place.
i just enjoy thinking about optimising scripts, so that's why i asked.

++ christoph ++

[user_deleted]

i also experienced a hdv nightmare once (together with under-exposed footage, camera motions without markers... you know... THE true nightmare, the director didn't follow any of my advise and thought he could shot with his sunday morning cam, although he paid the high-priced cyclo studio cyclo)
Anyone got experience with dvcpro-hd (4:2:2) new pana hvx200 cam ?
The price seems pretty low too, but i never had the chance to test chromakey footage coming from it

[user_deleted]

QUOTE(janimatic @ 03/06/06, 12:28 PM) [snapback]230092[/snapback]
Anyone got experience with dvcpro-hd (4:2:2) new pana hvx200 cam ?

i did some test with footage from the web...
seems that it's not that much better than well shot HDV as i had hoped... there's quite some noise and quite a bit of compression artifacts, and the resolution is not that high either. so you trade better color sampling vs other problems
the main advantage is that it's true progressive and the compression sheme doesnt break down on movements. works ok if you downscale to SD in the end (but this is true for HDV as well).

the best way to shoot HD greenscreen on a budget still seems to be to use the canon XL-H1 and capture over HD-SDI. total cost under 15'000 bucks. if you need a portable solution, the HVX is definitely worth a try.
++ christoph ++

[user_deleted]

thanks for the advise , so a cheap solution doesn't exist yet .... it's good to know

[user_deleted]

QUOTE(janimatic @ 03/06/06, 01:10 PM) [snapback]230096[/snapback]
thanks for the advise , so a cheap solution doesn't exist yet ....

well, i consider 15000 for a full uncompressed HD solution to be pretty cheap. specially because you would usually rent the camera, and need an editing machine anyway. so all you really need to buy is a HD-SDI capture card (600USD) and some fast drives (2000USD).
the only annoying bit is that it's still interlaced... so getting a HVX and an component HD card instead is tempting (but not as clean).
++ christoph ++

[user_deleted]

its really simple - you get what you pay for. Some cameras might have 4:2:2 color sampling but they still mpeg the footage making it an even worse candidate for VFX work than DV.