Coding mode decision algorithm for fast HEVC transrating using heuristics and machine learning

This article describes a framework to speed up the HEVC encoding decisions for on-demand transrating of bitstreams. The methods proposed collect information from a high-quality reference bitstream which after processing is used to limit the number of modes evaluated in subsequent re-encodings at different bitrates. In this way, the time required to process re-encode-time computing-intensive decisions, such as partitioning and motion estimation is significantly reduced. The methods proposed are a combination of heuristics with a statistical basis and fast decision techniques trained using automatic learning methodologies. Experimental results using the HEVC reference encoder show that jointly the methods proposed reduce the transcoding computational complexity by up to 78.8%, with Bjontegaard bitrate deltas penalties smaller than 1.06%. A comparison with related works showed that the proposed method is able to outperform state-of-the-art solutions in terms of combined rate-distortion-complexity performance indicators.