AN EFFICIENT COMPLEXITY REDUCTION ALGORITHM FOR CU SIZE DECISION IN HEVC

摘要

In brief, this work focuses on reducing the encoding complexity with less than 1% encoding efficiency loss for low delay (LD) and random access (RA) profiles in HEVC. An efficient CU (coding unit) size decision algorithm based on probabilistic graphical models is proposed for HEVC inter coding, which contains two methods: the CU size early termination (CUET) decision method and the CU size early skip (CUES) decision method. The CU pruning is modeled as a binary classification problem based on the Naive Bayes (NB) model Furthermore, a Markov random fields (MRF) model based method is presented to improve the algorithm performance. The difference from previous works is that residual flag in inter-coded CU and the neighboring information are used to determine the CU size. The offline learning method is used to obtain the statistical parameters. This presented approach can significantly reduce the encoding complexity. Furthermore, it can bring lower power cost for hardware implementation. The simulation experiment results demonstrate that this method can significantly reduce by 50.59% and 53.86% average encoding complexity under low delay and random access profiles, while the encoding efficiency can be reduced by 0.82% and 0.98% on average. Moreover, the rate-distortion (RD) performance of this method is nearly the same as HEVC reference software.