Coding unit complexity-based predictions of coding unit depth and prediction unit mode for efficient HEVC-to-SHVC transcoding with quality scalability

Fast Depth and Inter Mode Prediction for Quality Scalable High Efficiency Video Coding

“…The proposed algorithm is developed for RA coding structure. The QPs in the BL are set to (26,30,34,38), and the corresponding QPs in the ELs are set to (22,26,30,34) and (20,24,28,32), respectively. We have used the eight training sequences (Sunflower, Tractor, Flowervase, PartyScene, BlowingBubbles, RaceHorses, Parkruner and Town) in the four proposed strategies: CBDP, DB-IMMD, EVBNSMP and CDB-DET.…”

“…As "DB-IMMD", "RCB-SMET", and "EVB-NSMP" are related to mode, we also integrate these three strategies in our experiments, and referred to it as "Mode". We have tested both "Depth" and "Mode" for the EL with (22,26,30,34). We adopt BDBR to evaluate the coding efficiency loss.…”

“…For fair comparisons, all these three algorithms are implemented on the same computing platform. Two settings of QPs: Q1 = (22,26,30,34) and Q2 = (20,24,28,32) are used. The overall performance of our algorithm and Hamid's algorithm are listed in Table VI and Table VII, with Q1 and Q2 respectively.…”

“…Afterwards, if depth early termination is satisfied, other depth levels can be directly skipped to improve the coding speed. Yeh et al [26] propose a coding unit complexity (CUC)based prediction method to improve the coding speed of HEVC-to-SHVC transcoding for quality scalability. Yeh et al [27] use encoded Coding Unit (CU) sizes, prediction modes, motion vectors and Rate-Distortion Costs (RD-Costs) in BL and encoded CU sizes in EL to accelerate the coding speed of quality SHVC.…”

“…Therefore, we can use inter-layer and spatial correlations to predict candidate depths and skip depths with low likelihood to improve the coding speed. The relative CUs of the current CU, which are used to predict candidate depths of the current CU, are shown in Fig.2 (26,30,34,38), and the corresponding QPs used for the EL are (22,26,30,34). Based on these experiments, we propose our efficient fast decision methods below.…”