P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML
Video ̇Compression ̇OLE WL040/Bidgolio-Vol I WL040-Sample.cls September 14, 2003 18:10 Char Count= 0
540 VIDEOCOMPRESSIONincluding the chromaticity coordinates diagram defined
in a joint ISO/CIE standard ISO/CIE 10527:1991, which
supersedes CIE S002(1986). The range (gamut) of colors
visible to the human eye is described by the LAB gamut
in the CIE chromaticity diagram, larger than the EKTA
film gamut which is, in turn, larger than the RGB digital
media gamut.
Poynton (1999a,b) described aspects of color speci-
fication and coding relevant to video signal processing
in the form of Frequently Asked Questions available on-
line. The CIE home page (Commission Internationale
de l’Eclairage, 2000), the HyperPhysics Web site (Nave,
2001), and the Color Metric Converter Web pages (Dawes,
1999) serve to encourage further online exploration of
human vision and color concepts.
The analog NTSC television standard SMPTE 170M-
1999 specifies the system reference white and the pri-
mary color channels (green, blue, red) in terms of chro-
maticity coordinates The system reference white has
been chosen to match the chromaticity coordinates of
the D 65 illuminant defined by joint ISO/CIE standards
ISO/CIE ISO 10526:1999 and CIE S 005 E-1998 intended
to represent average daylight with a correlated color tem-
perature of approximately 6500K.
According to SMPTE 170M-1999 section 5, a cath-
ode ray tube display has an inherently nonlinear electro-
optical transfer characteristic and to achieve an overall
system characteristic that is linear, it is necessary to spec-
ify compensating nonlinearity elsewhere in the system.
In NTSC, PAL, and SECAM systems, the signal is
pre-corrected for non-linearity at the signal source as-
suming the display is intended to be viewed by human
observers in a dimly-lit environment. Although Table 1 of
ITU-R BT.470–6 (conventional television systems) lists the
assumed gamma of the display device for which a pre-
correction of a monochrome signal is made to be typi-
cally 2.2 (in some countries 2.8), the operating values of
the respective transfer characteristics may vary from
the precise values given in sections 5.1 and 5.2 of the
standard to meet operational requirements in practical
systems.
ITU-R BT.601–5 defines a method in section 3.5 for
constructing digital video luminance and color difference
signals (Y,CR,CB) comprising the steps of construct-
ing analog luminance EYand color difference signals
(ER′−EY,EB′−EY′) as weighted sums and differences of
the corresponding gamma-corrected analog television
color signals (ER′,EB′,EG′), renormalizing the analog televi-
sion luminance and color difference signals to the rangesEY′∈[1.0, 0.0], (E′R−EY′)∈[+ 0 .5,− 0 .5],(EB′−EY′)∈[+ 0 .5,− 0 .5],and quantizing these renormalized analog luminance and
color differences signals (chrominance) to a uniformly-
quantized 8-bit binary encoding equivalent to a decimal
range [0, 255] to provide digital luminance and color dif-
ference signals (Y,CR,CB). The digital luminance occu-
pies only 220 levels to provide working margins with black
at level 16 and the color difference signals occupy 255
levels with zero value at level 128. It is usual to limitthe gamut of digitally coded (Y,CR,CB) signal value to
that supported by the corresponding ranges of (R,G,B)
signals.Sampling, Quantizing, and Coding
We conventionally think of images as sampled and quan-
tized at a sampling rate large enough to preserve the use-
ful information. When we sample light measurements,
we usually record both the brightness (luminance) and
color (chrominance) information associated with an im-
age pixel. In a motion picture application we often record
the frame number (or time code) to indicate the position
or presentation time of the frame relative to the start of
the scene or video image sequence. The frame aspect ratio
of a rectangular image frame is the ratio of its width to
height.
International Telecommunications Union Recommen-
dation ITU-R BT.601–5 (formerly known as CCIR 601)
specifies methods for digitally coding standard 4:3 and
wide-screen 16:9 television video signals for both 525-line
and 625-line television systems. In particular, thepixel as-
pect ratio(defined as the ratio of pixel height to width)
may be more or less than 1/1 resulting in rectangular (non-
square) rather than square pixels. Aho (2002) showed how
to deduce the ratio 11/10 for rendering data from 525-line
systems and 54/59 for 625-line systems from analog stan-
dard video sampling rates.
ITU BT.601–5 recommends co-siting the samples
representing digital luminance and color difference
signals (Y,CR,CB) to facilitate the processing of digital
component video signals and further recommends sub-
sampling to reduce the data rate. Horizontal subsampling
of the color difference signals by 2:1 yields a 2/3 saving
in data rate over (R,G,B) with almost imperceptible
change in visual quality when implemented with proper
decimation and interpolation filters. The 4:2:2 nomen-
clature indicates 4 luminance samples for every 2 color
difference samples in a scan line. Horizontal and vertical
subsampling by 2:1 is denoted by the 4:2:0 and yields
about 1/2 the data rate for (R,G,B). In turn, 4:1:1 denotes
horizontal subsampling by a factor of 4 without vertical
subsampling. If a fourth parameter is mentioned—as in
4:4:4:4—it refers to the alpha channel required for keying
applications. Note that digital luminance levels 0 and
255 are reserved for synchronization in 4:2:2 systems
while levels 1 to 254 are available for video. Figure 2a
illustrates the digital sampling of luminance and chromi-
nance. Each cell indicates a luminance sample while
chrominance samples are indicated by cells designated
by “C” and alpha channels by cells designated by “A.”
The 4:4:4:4 sampling scheme samples chrominance
and alpha channels with every luminance sample. The
4:2:2, 4:1:1, and 4:4:4:4 schemes sample chrominance
with luminance samples while the 4:2:0 scheme samples
chrominance between luminance samples.
ITU-R BT.601–5 digital video is sampled at 13.5 million
pixels per second (for both 525- and 625-line systems). A
525-line analog NTSC (ANSI/SMPTE 170M-1994) video
signal is sampled at exactly 12+ 27 /99 million pixels per
second when sampling with square pixels. A 625-line
analog PAL (ITU-R BT.470–3) video signal is sampled at