Video Compression - Signal Processing Systems - page 58

Best
Matching
Block

Current
MB

Frame

Ref-4
(t-4)

Ref-3
(t-3)

Ref-1
(t-1)

Current
(t)

Ref-2
(t-2)

Fig. 12 Illustration of Multiple Reference Frame (MRF) scheme

For intra (spatial) prediction, there are two modes defined in H.264/AVC
Baseline Profile as introduced in Sect. 2.2 . Intra4

×

4 mode is suitable for regions

×

with detailed texture. On the contrary, Intra 16 16 mode is suitable for regions

with smooth texture. In H.264/AVC High Profile, anther intra prediction mode Intra
8

×

8 is added for the texture with the medium size. Each MB is partitioned into

×

8 8 blocks. There are also nine modes provided and the predictive directions are

the same with those of Intra 4

×

4 modes.
For entropy coding, H.264/AVC adopts Context-based Adaptive Variable Length
Coding (CAVLC) and Context-based Adaptive Binary Arithmetic Coding (CABAC)
[ 14 ] . Compared to the previous standards, CAVLC and CABAC adaptively adjust
probability models according to the context information. Accurate probability
models lead to better coding performance for entropy coding. In addition, CABAC
can save 10% bitrate compared to CAVLC. It is because arithmetic coding can
achieve better coding performance than variable length coding. There are still some
other coding tools in H.264/AVC High Profile like Adaptive Transform Block-size
and Perceptual Quantization Scaling Matrices. Please refer to [ 13 , 21 ] for the details.

3.2 Streaming and Surveillance Applications and Scalable

Video Coding

There are various of multimedia platforms with different communication channels
and display sizes. The variety of multimedia systems makes scalability become
important for video standards to support various demands. Scalability of video
coding means one encoded bitstream can be partially or fully decoded to generate
several kinds of videos. More decoding data contribute higher spatial/temporal
resolution or better visual quality. For a streaming service, its clients may be a
mobile phone, a PC, or an HD TV. Without scalability, we need to encode a
video into different bitstreams for different requirements. It is not efficient for
storage. Another key application for scalable video coding is surveillance. Due to
the limited storage size, less important data should be removed from the bitstream.
The importance of the information in a surveillance video is decayed with time. We