We introduce a new parallelization method for high-efficiency video coding (HEVC), which resolves the shortcomings of the existing tile-based parallel processing method. The parallel HEVC performs encoding by dividing a frame into numerous parallel units. This decreases the compression efficiency compared with sequential HEVC, because it artificially breaks the data correlation within a frame, which is called the parallelization overhead. The traditional parallel techniques such as Tiles and wavefront parallel processing (WPP) inherently introduce a high parallelization overhead because they simply divide a frame statically without considering the contents of the frame. The proposed new parallel encoding scheme resolves such problems by partitioning a frame based on the meaningful contents. In order to analyze the correlations within a frame and define the contents, the features within a frame are first extracted and clustered. In the feature clustering algorithm, two factors are considered to balance the workload between parallel units: (1) the number of features in each cluster and (2) the number of coding tree units (CTU) occupied by each cluster. The frame is partitioned based on the result of clustering, and the partitions are encoded in parallel. The proposed scheme achieves a bit-saving of up to 7.21%, with an average of 3.71%, along with an average time-saving of 20.50% compared to the Tiles technique.
|Number of pages||16|
|Journal||Multimedia Tools and Applications|
|Publication status||Published - 2019 May 1|
Bibliographical noteFunding Information:
Acknowledgements This work was funded by grants from the Digital Media & Communication R&D Team, Samsung Electronics Co., Ltd. and by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP)(No. NRF-2018R1A2A2A05018941). Won Woo Ro is the corresponding author.
This work was funded by grants from the Digital Media & Communication R&D Team, Samsung Electronics Co., Ltd. and by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP)(No. NRF-2018R1A2A2A05018941). Won Woo Ro is the corresponding author.
© 2018, Springer Science+Business Media, LLC, part of Springer Nature.
All Science Journal Classification (ASJC) codes
- Media Technology
- Hardware and Architecture
- Computer Networks and Communications