Title :
Fast mode decision algorithm based on thread-level parallelization and thread slipstreaming in H.264 video coding
Author_Institution :
Multimedia Platform Lab, DMC R&D Center, DMC Business, Samsung Electronics
Abstract :
As multicore processors have become a general in computer architecture, parallel algorithms and software designs based on thread-level parallelization are rapidly gaining importance in compute-intensive applications. However, it is difficult to exploit thread-level parallelism for multimedia applications like video encoders due to complicated data/control dependences. In particular, the H.264 encoder has data dependences between the inter mode decision and the intra mode decision, especially when rate-distortion optimization (RDO) is used. This paper proposes a new RDO mode decision algorithm based on thread-level parallelization for the H.264 encoder, which can efficiently resolve the dependences and exploit thread-level parallelism for fast mode decision. The proposed methods, slipstreaming and skipping mechanisms reduce the total encoding time without PSNR loss and bit rate increment. These mechanisms can remove most of computation overhead for RDO intra mode decision as well as improve the computational performance of motion estimation. The experiments performed with JM12.2 showed that the proposed method reduces the total encoding time by 31.3% on average while maintaining the identical PSNR and bit rate.
Keywords :
motion estimation; optimisation; rate distortion theory; video coding; H.264 video coding; computer architecture; fast mode decision algorithm; motion estimation; multicore processor; parallel algorithm; rate distortion optimization; software design; thread level parallelization; video encoder; Bit rate; Entropy coding; Instruction sets; Multicore processing; PSNR; H.264; RDO; mode decision; multicore; thread-level parallelization; video coding;
Conference_Titel :
Multimedia and Expo (ICME), 2010 IEEE International Conference on
Conference_Location :
Suntec City
Print_ISBN :
978-1-4244-7491-2
DOI :
10.1109/ICME.2010.5582556