HEVC Parameter Exploration for Efficient Mode Decision
Permanent address of the item is
Parametrien etsintä HEVC:n tehokkaalle moodivalinnalle
High Efficiency Video Coding (HEVC) is the latest video coding standard. It halves the achieved bit rate compared with the previous standard, Advanced Video Coding (AVC). However, the bit rate decrease comes with 40% increase in encoding complexity. This is mainly due to larger number of block coding modes, including Symmetric motion partitions (SMPs), Asymmetric motion partitions (AMPs), and larger coding units of up to 64x64 pixels. These new features are mainly used for Inter prediction that accounts for 60-70% of the whole encoding time. For this reason, optimization of Inter prediction is the main topic in this Thesis. To tackle the Inter prediction complexity, a parametric exploration was chosen as the approach. The exploration was done by gradually shifting the focus from the most coarse optimization to the parameter fine tuning. The selected approach in this study required thousands of individual tests so an automated solution was needed. This led to the creation of a new software solution, TUT Task Manager. It is capable of automatically distributing the tasks of parametric exploration to any number of nodes available in the local network. In total, TUT Task Manager was used to run 4000 tests with a combined CPU time of 14 months. The results were used to create a set of recommended schemes for Inter mode selection. Overall, these new schemes are shown to provide 31-50% complexity saving against the default configuration of HM 11.0, with a minor bit rate increase of 0.2-1.3%. They also provide better RDC performance than the existing solutions. The tools and methods used in this work are so generic that they can be used to further optimize other parts of the video codec.