Search range reduction for uni-prediction and bi-prediction in HEVC

Abstract

The motion estimation block of the high-efficiency video coding (HEVC) standard is highly complex. This paper proposes two algorithms, namely external search range reduction (ESR) and internal search range reduction (ISR) to reduce the motion estimation complexity of HEVC in uni-prediction and bi-prediction both in the fast search mode. The proposed ESR algorithm for uni-prediction motion estimation reduces the search range adaptively for Test Zonal Search (TZS) motion estimation with negligible loss in coding efficiency. The proposed ISR algorithm for uni-prediction, namely ISR-R, limits the search range further in the raster search stage of the TZS algorithm. Moreover, a fast bi-prediction motion estimation algorithm is proposed which includes both ESR and ISR to reduce the motion estimation complexity in bi-prediction. Our algorithms are implemented in the HM-16.6 encoder in the fast search mode. The performance of the proposed algorithms are tested individually and then by combining all the algorithms. When the algorithms are combined, the number of search points and motion estimation time are reduced by 60.58 and 59.58 %, respectively in the fast search mode with a BD-Rate of 0.193 % and BD-PSNR of −0.005 % in the LD-B main profile. The number of search points and motion estimation time are reduced by 59.55 and 57.71 %, respectively in the fast search mode with a BD-Rate of 0.265 % and BD-PSNR of −0.008 % in the RA main profile.