Improved reference picture list sorting in video coding
Sebastian Schwarz, Marta Mrak · 2015
Motion compensated prediction using multiple references is a key element of any modern video coding standard. A smart and meaningful selection of references contributes significantly to encoding efficiency. This selection could either be performed through computationally expensive pre-analysis and reference picture selection optimisation, or through improved reference picture list structures based on general encoder decisions. This paper analyses encoder reference picture decisions for a maximum set of available reference pictures. Characteristic encoder decision properties are identified, then two variants of the standard HEVC reference picture list structure and sorting approach are derived and implemented. Evaluation verifies that the conclusions drawn from the general encoder decisions hold. The proposed changes provide coding efficiency benefits in terms of bit rate savings of up to 6%, with a limited increase in computational complexity of around 11%.