Improving HEVC Coding Efficiency Using Virtual Long-Term Reference Pictures

Inter-frame prediction in HEVC uses two types of reference pictures: short-term and long-term. Out of these, long-term reference (LTR) pictures enable exploiting correlation among frames with extended temporal distances. In addition, LTR pictures improve the inter-frame prediction where video scenes are repeated such as in TV-series episodes, news broadcasts and movies. In this context, this paper proposes an algorithm to calculate LTR pictures using artificially generated virtual reference frames for static-camera scenes. The experimental results demonstrate an average coding improvement of 2.34% in terms of Bjøntegaard Delta Bit Rate(BDBR), when compared with the HEVC reference encoder HM16.8.