ALSO SEE: BIT
- or “Chroma Subsampling”
- A means of reducing the amount of data required to store or transmit a video picture by sampling CHROMA less frequently than LUMA.
Cinematography is about creating beautiful moving pictures, and that means storing immense amounts of information about the sequence of individual still images that form the movie clip. When compared with stills imaging, the importance of using the available storage capacity as efficiently as possible is much greater in moving picture systems due to the sheer number of individual “frames” involved in the process. At 24 frames per second, a 60 minute film will display 86,400 single frames – that’s a big bill at snappy snaps.
Subsampling is a process used in both analogue and digital camera systems, and CODECs, in order to get the most out of a limited DATA RATE. It relies on the human eye’s tendency to be less responsive to CHROMA information than it is to LUMA. This is due to the eye having a high proportion of “rods” (which are highly sensitive to LUMA but play little part in resolving CHROMA) to the CHROMA sensitive “cones”. The result is that when sampling an image it is possible to record less CHROMA data than LUMA data without necessarily reducing the perceived quality of the resulting image.
Subsampling works on a given array of photosites or pixels by taking fewer CHROMA samples than LUMA samples. The nomenclature used to describe different types of SUBSAMPLING takes the form:
J = number of horizontal luma (Y) samples
a = the number of CrCb “colour pair” (U+V) samples on the first line
b = the number of CrCb “colour pair” (U+V) samples on the second line
*Please note: This nomenclature has evolved in a haphazard fashion and can be confusing! J originally described the LUMA sampling as a multiple of 3 3/8 MHz, and the “b” figure described the Cr subsampling only. HDTV would have been shown as 22:11:11. The nomenclature has been simplified to reflect the systems generally in use and, as described here, is used by manufacturers and developers to roughly describe those systems. It cannot accurately describe all subsampling systems and has particular problems with vertical subsampling.
The nomenclature refers to subsampling applied to an 8 PIXEL or PHOTOSITE array of 4 across by 2 down. Each of the 8 units can potentially have a sample for each of LUMA, Cb and Cr, giving 24 “units” of information. In the nomenclature “a” and “b” refer to CbCr pairs, which have twice the amount of information as the LUMA samples referred to by “J”.
- 4:4:4 or no sampling – each line has full LUMA sampling (8 units of data), the first line has full CrCb sampling (8 units of data), the second line has full CrCb sampling (8 units of data). The full picture – 24 units of data. (Used in DUAL-LINK SDI).
- 4:2:2 subsampling – each line has full LUMA sampling (8 units of data), the first line has 2 CrCb subsamples (4 units of data), the second line has 2 CrCb subsamples (4 units of data). Total units of data is 16 so a 33% saving compared to 4:4:4. (Used in Digi-Beta, XDCAM and ProRes amongst others).
- 4:2:0 subsampling – each line has full LUMA sampling (8 units of data), the first line has 2 CrCb subsamples (4 units of ata), the second line has no colour subsampling at all. Total units of data is 12 representing a 50% saving compared to 4:4:4. (Used in MPEG CODECs, common JPEG settings, and PAL DVCAM amongst others).
- 4:1:1 subsampling – each line has full LUMA sampling (8 units of data), the second line has 1 CbCr subsample (2 units of data), the second line has 1 CbCr subsample (2 units of data). Total units of data is also 12 representing a 50% saving compared to 4:4:4. (Used in DVC Pro and NTSC DVCAM amongst others).
Where there is a fourth figure in the ratio i.e. J:a:b:x most often in 4:4:4:4, the last figure refers to the ALPHA channel, a measure of transparency used in CGI.
There are many details specific to proprietary subsampling systems which are not expressed by this nomenclature, including at which PIXELS or PHOTOSITES the subsampling takes place or whether an average is taken across more than one; the timing of the subsampling, particularly with regard to interlaced signals, and so on. These can all impact on the suitability of different systems for different tasks.
Subsampling is a form of LOSSY COMPRESSION and visual information is therefore lost. It is ubiquitous and used in most forms of displayed video systems and all but the highest end camera systems. While the result may be imperceptible when used correctly there is always a danger of losing too much visual information and the potential to generate ARTEFACTS. Especially so when converting from one subsampled format to another.
ALSO SEE: BIT