|
|
 |
These comments are
originally from John Carey (taken from an email and a newsgroup post). I've
edited them down a bit, though the content should still be accurate. John
has worked many years in the film and FX industry and was, the last I heard,
working at Disney.
"The problem with doing color difference matting like chromakey and Ultimatte
with DV doesn't arise from the CoDec, but rather from the sampling scheme.
The video image is sampled prior to the codec and undersampling is the
beginning of the DV bitrate reduction scheme, but it happens prior to encoding.
In other words, there is no digital information to compress prior to sampling.
DV uses either 4:1:1 or 4:2:0 depending on whether or not it is 525/30 or
625/25. That means that while the luminance is being fully sampled for each
pixel because it is being sampled at 4 times the color subcarrier, the
chrominance is only being fully sampled at the subcarrier rate which means that
you only get one color sample for every 4th pixel. This happens prior to the
compression step, which I believe is Huffman, not RLE.
Since you only have one color sample for each 4th pixel, that means that your
color difference mattes which are generated based on chrominance have only 1/4
of the resolving power of your luminance channel and a composite made from them
will result in highly aliased edges. That means that no new advances in the
CoDec are going to make the DV images any better for effects work.
The combination of undersampling and the codec don't add the objectionable
artifacts to which you refer. The bitrate reduction employed in DV is designed
to keep nearly all of the perceptually important information and to eliminate
redundant or perceptually unimportant information. It happens that the human
eye is capable of perceiving significantly more contrast information than it is
color information. For that reason, the fact that DV only has color information
on every fourth pixel doesn't seem objectionable. When you create a matte
signal from color difference, like Ultimatte or ChromaKey's do, you take color
difference information and make a contrast mask, or matte out of it, thereby
taking low resolution color information and deriving luminance information from
it and the human eye is far more sensitive to that."
BACK
|
|
|