Graphics Reference
In-Depth Information
Tiles are also well-suited for a general parallelization of encoder and decoder.
The amount of parallelism is not fixed, as is the case for WPP, allowing the encoder
to adjust the number of tiles according to its computational resources. Because
tiles can be used to subdivide the picture into multiple rectangles spanning the
picture horizontally and vertically, they are also better suited for region of interest
(ROI) coding than, e.g., slices. In conversational applications, for example, tiles
in combination with a tracking algorithm can be used to dynamically adjust the
size and error protection of the ROIs. Tiles are also good for applications where
coding tasks need to be distributed onto different hardware machines/systems, since
each tile can be, assuming cross tile-border filtering is done separately, processed
relatively independent from other tiles within the picture and therefore, in this case
require a minimum of communication between the threads processing the individual
tiles. However, tiles usually come along with the cost of a significant overhead in
bit rate.
In order to simplify implementations of the standard, HEVC does not allow the
use of tiles and WPP simultaneously in the same compressed video sequence. It
may be interesting, however, to allow some combination of these tools in the future.
For instance, it could be necessary to divide an ultra-high definition video signal
into sub-pictures using tiles with WPP inside each sub-picture, to enable real-time
encoding/decoding.
3.4
Conclusions
The nested quadtree-based block partitioning as a distinguished feature of HEVC
has been presented. It has been shown that more than half of the coding efficiency
improvements of HEVC relative to H.264 j MPEG-4 AVC for HD video material is
obtained solely by introducing this flexible block partitioning concept for improved
prediction and transform coding. Furthermore, it has been elaborated on how the
two novel picture partitioning features of tiles and wavefront parallel processing can
improve the parallel-processing friendliness of HEVC in order to meet the increased
demand in computational complexity. Finally, it has also been discussed how the
new concept of dependent slice segments in HEVC can facilitate ultra-low delay
processing on a sub-picture level without substantial loss in coding efficiency, when
used together with the aforementioned novel picture partitioning tools.
References
1. Ahmed N, Natarajan T, Rao KR (1974) Discrete cosine transform. IEEE Trans Comput
C-23:90-93
2. Alvarez-Mesa M, George V, Chi CC, Schierl T (2012) Improving parallelization efficiency
of WPP using overlapped wavefront, Joint Collaborative Team on Video Coding (JCT-VC),
Document JCTVC-J0425, Stockholm, July 2012
Search WWH ::




Custom Search