Information Technology Reference
In-Depth Information
determining the relative place of a frame in a sequence, for editing purposes and
for synchronizing video and audio. Usage of identical time codes implies that cur-
rent production systems can be easily adapted from SD to HD.
3 H.264/AVC
H.264/AVC is a highly efficient video compression standard which was created
jointly by ITU-T and ISO/IEC. The initial H.264/AVC standard was finalized in
2003 and focused on video material with 4:2:0 chroma subsampling and a bit-
depth of 8 bits [8]. In a later stage, extensions to support higher resolutions and
bit-depths and different chroma subsampling options were added under the name
FRExt (Fidelity Range Extensions). These extensions include an 8x8 DCT trans-
form and intra prediction, support for adaptive transform sizes, custom quantiza-
tion matrices, lossless coding, support for 4:2:2 and 4:4:4 chroma subsampling
and support for higher bit-depths [9].
Mode information, motion vectors
Di vi d e i n t o
MBs
NAL units
Pr e d i c t i o n
Tr a n sf o r m
Quantization
CABAC/ CAVLC
Mode
dec.
Inverse
Quantization
ME/ MC
Int ra-predict ion
Refer ence fr ame
buffer
(L0 and L1)
Inverse
Tr an sf o r m
Reconstructed
current frame
Deblocking
filter
Fig. 1
H.264/AVC encoder architecture
Figure 1 illustrates H.264/AVC's classical block-based hybrid coding architec-
ture. Each picture is divided into one or more slices, each consisting of a number
of macroblocks. A macroblock (MB) covers a rectangular picture area of 16x16
luma samples and, corresponding to this, two rectangular areas of chroma samples
(when 4:2:0 chroma subsampling is used, two blocks of 8x8 chroma samples).
Each MB is either spatially (intra) or temporally (inter) predicted and the predic-
tion residual is transform coded. H.264/AVC supports three basic slice coding
types: I slices, P slices and B slices. H.264/AVC Intra, which is used in our ex-
periments, only allows the use of I slices. I slices are coded without any reference
to previously coded pictures. Macroblocks in I slices are always spatially pre-
dicted using directional intra-prediction. Intra-prediction can be performed on
blocks of 4x4, 8x8 and 16x16 samples. Each block is predicted based on
neighbouring sample values as illustrated in Figure 2 for 4x4 and 16x16 blocks.
The directional 8x8 intra prediction modes introduced in FRExt are similar in de-
sign to their 4x4 equivalents [8, 9]. Macroblocks in P and B slices can additionally
