DocumentCode :
610115
Title :
A High Throughput Multi Symbol CABAC Framework for Hybrid Video Codecs
Author :
Rapaka, K. ; En-Hui Yang
Author_Institution :
Univ. of Waterloo, Waterloo, ON, Canada
fYear :
2013
fDate :
20-22 March 2013
Firstpage :
515
Lastpage :
515
Abstract :
Summary form only given. This paper proposes a Multi-Symbol Context Adaptive Binary Arithmetic Coding (CABAC) Framework in Hybrid Video Coding. Advanced CABAC techniques have been employed in popular video coding technologies like H264-AVC, HEVC. The proposed framework aims at extending these technique by providing symbol level scalability in being able to code one or multi-symbols at a time without changing the existing framework. Such a coding not only can exploit higher order statistical dependencies on a syntax element level but also reduce the number of coded bins. New syntax elements and their Probability modeling are proposed as extensions to achieve Multi-Symbol coding. An example variant of this framework, that is coding only maximum of two symbols at a time for quantized coefficient Indices, was implemented on top of JM18.3-H264 CABAC. This example extension when tested with on HEVC test Sequences shows significant throughput improvement (i.e., significant reduction in number of bins to be coded) and at the same time reduces Bit-rate significantly. The Frame-work can be seamlessly extended to code Multiple Symbols greater than two.
Keywords :
data compression; higher order statistics; image sequences; probability; video codecs; video coding; H264-AVC; HEVC test sequences; JM18.3-H264 CABAC; coded bin number reduction; high throughput multisymbol CABAC framework; high throughput multisymbol context adaptive binary arithmetic coding framework; higher order statistical dependencies; hybrid video codecs; hybrid video coding; probability modeling; Data compression; CABAC; Entropy Coding; H264; Hybrid Video codecs; curve fitting;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Data Compression Conference (DCC), 2013
Conference_Location :
Snowbird, UT
ISSN :
1068-0314
Print_ISBN :
978-1-4673-6037-1
Type :
conf
DOI :
10.1109/DCC.2013.94
Filename :
6543125
Link To Document :
بازگشت