Graphics Reference
In-Depth Information
To keep each slice within this limit and, at the same time, minimize the
packetization overhead, the encoder may produce slices with varying number
of CTUs within a picture.
3.
Parallel Processing
: To partition the picture into units which can be processed
in parallel. This is given by the fact that all slice-based encoding/decoding
operations including reconstruction prior to loop filtering can be independently
carried out in parallel.
SimilartowhatisspecifiedinH.264
j
MPEG-4 AVC, slices in HEVC consist of
an integer number of its minimum building block which, as already noted above,
is given by a CTU instead of a macroblock. CTUs in a slice are processed in
raster scan order such that each slice of a picture is independently parsable and
decodable. This is achieved by terminating the CABAC bitstream at the end of
each slice and by breaking CTU dependencies across slice boundaries within a
picture such as, e.g., dependencies used for in-picture prediction, context selection,
or probability estimation. Due to this reduced exploitation of spatial redundancy, the
coding efficiency usually decreases quite substantially with increasing the number
of slices used for a picture.
Conceptually, a slice consists of a slice header and the slice data. The slice header
provides specific information for the decoding of the slice data, i.e., the coded CTUs
within the picture to which the slice belongs. Therefore, the slice header precedes
the actual slice data. Note that the overhead of each slice header also contributes
to the reduced coding efficiency when using multiple slices, especially at lower bit
rates.
During the development of HEVC it turned out that the conventional slice
concept, as outlined above and supported by prior video coding standards, is a
too rigid concept to properly meet all anticipated needs. In particular, the bit-rate
overhead caused by multiple slice headers and the strict breaking of in-picture
dependencies at slice boundaries was found to be critical in certain application use
cases.
Therefore, HEVC introduces the novel functionality of slice fragmentation at
conceptually two distinct levels. For the first level of fragmentation, each slice can
be divided into one or more slice segments at the CTU boundaries. The first slice
segment of a slice (in CTU raster scan order) is the independent slice segment and
includes the full slice (segment) header. The independent slice segment is also often
referred to as a regular slice, since it is conceptually equivalent to what is specified
in H.264
j
MPEG-4 AVC. All subsequent slice segments within a slice (if any) are
so-called dependent slice segments with drastically shortened slice segment headers.
Note that within the same slice CTU dependencies across slice segment boundaries,
both in terms of in-picture prediction and entropy coding, are allowed for dependent
slice segments, provided that no further restrictions are given. Slice segments will
be discussed in more detail in Sect.
3.3.1.1
below.
The second slice fragmentation level of HEVC is given by the instrument of
slice segment subsets, often referred to as substreams. Each slice segment subset
contains all coded bits of a subset of CTUs covered by the corresponding slice
