|
AFAIK all the MPEG2 encoders used for SD programs in the UK use 8 bits per component. The HD MPEG4/H264 encoders may use 10 bits rather than 8 but I have no evidence. The improvement when moving from 8 to 10 bits is very marginal on real pictures. It can of course be seen on special test signals such as shallow ramps.
PAL is no longer used in the UK and quite likely not anywhere else in Europe. Any material originated in PAL would be decoded to components before being used. Even an inexpensive single chip decoder can work remarkably well, leaving very few artifacts. The BBC and others have developed sophisticated 3D comb decoders that can convert PAL to component with essentially no artifacts. Quite a few people say "PAL" when the really mean 625 line, 50Hz. Just as some say NTSC when they mean 525 line, 59.94Hz. Likewise a lot of people say YUV when they really mean YCbCr or YPbPr. U and V are strictly speaking the weighted colour difference components immediately before being modulated on to a colour subcarrier. U and V have no place in a component system.
I cannot believe that anyone anywhere is using 3 bits per component at the input to the encoder. The pictures would be unbelievably and utterly impossibly bad.
Many years ago I experimentally digitised composite PAL video at 1 bit per pixel. As you would expect the pictures (in monochrome) looked truly horrible. The surprise was that in highly coloured areas the PAL subcarrier acted as a dither signal and made the picture vastly better. This was a demonstration of how you can trade quantising errors (limited bit depth) for noise. You can make a 3 bits per component picture without the posterisation errors you would expect simply by adding enough random noise. It will then look like a snowstorm.
|