Fast Coding Unit Selection Based on Local Texture Characteristics for HEVC Intra Frame

Miloš Radosavljević, Georgios Georgakarakos, Sebastien Lafond, Dejan Vukobratović

Forskningsoutput: Kapitel i bok/konferenshandlingKonferensbidragVetenskapligPeer review

Sammanfattning

High Efficiency Video Coding (HEVC) is a novel video compression standard. It provides significantly better per- formance than its predecessor, but introduces high computational complexity due its hierarchical block structure. Determining Coding Units (CU), Prediction Units (PU), and Transform Units (TU) size is a time consuming process, as block partitioning decisions are based on rate-distortion optimization (RDO). Tak- ing into account all possible outcomes in terms of different block sizes and different prediction modes that HEVC can exploit, RDO is computationally very intense. To speed-up the block partitioning decisions we propose a novel approach for the CU size selection for intra frames based on local image characteristics. We use data from the source image itself to make decisions on whether to split CU or not. Split decisions are based on the histogram matching of the Local Binary Patterns on two consecutive CU depths. Results show that encoding time can be reduced compared to the HM16.2 reference software by bypassing RDO calculation for some specific CUs. Different performance-complexity configurations are investigated. Based on the proposed configuration, performance loss varies from almost negligible up to 0.87dB using BD-PSNR metric. However, speed-up is from 5.4% up to 80.2% on average.
OriginalspråkOdefinierat/okänt
Titel på gästpublikation2015 IEEE Global Conference on Signal and Information Processing
RedaktörerMoura Jose, Oliver Wu Dapeng
FörlagIEEE Global Conference on Signal and Information Processing
Sidor1–5
ISBN (tryckt)978-1-4799-7590-7
StatusPublicerad - 2015
MoE-publikationstypA4 Artikel i en konferenspublikation
EvenemangIEEE Global Conference on Signal and Information Processing - IEEE Global Conference on Signal and Information Processing
Varaktighet: 14 dec 201516 dec 2015

Konferens

KonferensIEEE Global Conference on Signal and Information Processing
Period14/12/1516/12/15

Citera det här