Low complexity reference frame selection in QTBT structure for JVET future video coding

Abstract

In this paper, we propose a reference frame search method for JVET future video codec (FVC) that employs the quadtree plus binary tree (QTBT) structure. Among many new technologies proposed in FVC, QTBT poses a significant challenge since it contains the structural change of coding tree unit from HEVC. To reduce the encoding complexity of FVC with QTBT structure, we investigated some redundancy in motion estimation process-particularly, the reference frame search. In this paper, we present a method that effectively restricts the reference frame search range of general motion estimation as well as of affine motion estimation, exploiting the dependence within QTBT structure. The proposed method minimizes the maximum of the reference frame search ranges per each coding unit (CU) based on the prediction information of parents node. To be specific, the prediction direction and the index of reference frame of parent node were used. In addition, the proposed method utilizes the information of binary tree depth and of temporal layer to prevent undesired coding loss. The experimental results showed that the proposed method decreased the encoding time of motion estimation by 34% on average in comparison with joint exploration test model (JEM) 3.1, maintaining a reasonable coding efficiency (less than a 0.3% BD-rate loss).