Page 88 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 3.4. The H.263 Standard 65

layer to increase perceived quality by either increasing the picture rate (tem-
poral scalability), increasing the picture SNR quality (SNR scalability), or
increasing the picture resolution (spatial scalability). This mode has support
for three types of scalability: temporal, SNR, and spatial scalability, as de-
tailed next. This mode can be helpful when used over heterogenous networks
with varying bandwidth capacity and also in conjunction with error correction
schemes.
a: Temporal scalability: Temporal scalability refers to enhancement infor-
mation used to increase the picture quality by increasing the picture display
rate. Temporal scalability is achieved by employing bidirectionally predicted
pictures, or B-pictures. B-pictures can be predicted from a previous and=or
a subsequent reconstructed picture in the reference layer (the layer used for
prediction). B-pictures in this mode di1er from the B-picture part of a PB- (or
an Improved PB-) frame in that they are separate entities in the bitstream. In
other words, they are not syntactically intermixed with a subsequent P-picture.
It should be emphasised that B-pictures should not be used as reference pic-
tures for the prediction of any other picture. This is particularly important
to allow for B-pictures to be discarded if necessary without adversely a1ect-
ing any subsequent pictures, thus providing temporal scalability. Figure 3.6(a)
illustrates temporal scalability using B-pictures. It should be pointed out that
the location of B-pictures in the bitstream is in a data-dependence order rather
than in a temporal order. For example, in the case shown in Figure 3.6(a)
the bitstream order of the encoded pictures is I 1 ;P 3 ;B 2 ;P 5 ;B 4 ;::: : There
is no limit to the number of B-pictures that may be inserted between pairs
of reference pictures in the reference layer. In this mode, motion vectors are
allowed to extend beyond the picture boundaries of B-pictures.
b: SNR scalability: SNR scalability refers to enhancement information used
to increase the picture quality without increasing picture resolution. The pro-
cess of compression usually introduces artefacts and distortions. As a result,
the di1erence between a reconstructed picture and its original in the encoder is
almost always a nonzero-valued picture. Normally, this coding error picture is
lost at the encoder and never recovered. With SNR scalability, however, these
coding-error pictures can be encoded and sent to the decoder. At the decoder,
such coding-error pictures can be used to increase the signal-to-noise ratio of
the decoded picture, and hence the term SNR scalability. Figure 3.6(b) illus-
trates SNR scalability. If the enhancement-layer picture is predicted only from
a simultaneous lower-layer reference picture, then the enhancement-layer pic-
ture is referred to as an EI-picture. If, however, the enhancement-layer picture
is bidirectionally predicted using both a prior enhancement-layer picture and a
temporally simultaneous lower-layer reference picture, then the enhancement-
layer picture is referred to as an EP-picture. The picture in the reference