Page 37 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

14 Chapter 2. Video Coding: Fundamentals

The amplitude of the ltered analog signal is then sampled at speci c time
instants to generate a discrete-time signal. The minimum sampling rate is
known as the Nyquist rate and is equal to twice the signal bandwidth.
The resulting discrete-time samples have continuous amplitudes. Thus, it
would require in nite precision to represent them. The quantization operation
is used to map such values onto a nite set of discrete amplitudes that can be
represented by a nite number of bits.
Each discrete-time, discrete-amplitude sample is called a picture element
and is usually abbreviated to a pel or a pixel. The pels are arranged in a
two-dimensional (2-D) array to form a digital still image or a digital frame.
A digital video consists of a sequence of such digital frames.
For color video, the foregoing operations are repeated for each component.
Thus, a digital still image would normally be represented by three 2-D arrays.
Almost all digital video systems use component representation. This avoids
the artefacts that result from composite encoding. 9
As an example, consider the digitization of a 625=50 PAL analog signal.
The luma and chroma components are rst ltered to 5:5MHz and 1:5MHz,
respectively. During sampling, minimum sampling frequencies of 11 MHz and
3MHz must be used to sample the luma and chroma components, respectively.
The resulting discrete-time signals are then quantized to a given precision
(usually 8 bits).

2.4.3 Chroma Subsampling

As already mentioned, the HVS has poor response to chrominance spatial
detail compared to its response to luminance spatial detail. This property can
be exploited to reduce bandwidth requirements by subsampling the chroma
components. The most commonly used subsampling patterns are illustrated
in Figure 2.2. In 4:2:2 subsampling, the chroma components are subsampled
by a factor of 2 horizontally. This gives a reduction of about 33% in the
overall raw data rate. In 4:1:1 subsampling, the chroma components are sub-
sampled by a factor of 4 horizontally, giving a reduction of 50%. In 4:2:0
subsampling, the chroma components are subsampled by a factor of 2 both
horizontally and vertically, giving a reduction of 50% in the overall raw
data rate. Vertically subsampled chroma samples are always sited midway
between luma samples. Horizontally subsampled chroma samples, however,

9 As already discussed, composite encoding is used in analog systems to save bandwidth. In
digital systems, however, bandwidth is saved using digital video compression techniques, as will
be described later.