由於多視角視訊編碼運算複雜度極高，若需實現於消費電子產品中，須加速其執行，此可由演算法與硬體層進行。而為了讓快速演算法能夠適用於不同量化參數，本論文提出以運動向量差為基礎之快速模式決策演算法。本論文設計包含三個部分，由於最佳編碼模式的決定與RD cost有很高的關聯性且direct佔極高比例，因此第一部分演算法依據RD cost之統計分析針對direct模式進行加速，找出決定候選模式之臨界值與QP之間呈指數函數的關聯性，加速時針對不同QP的臨界值，可直接從指數函數中求得，並將RD cost與臨界值比較以減少測試其他Inter modes及Intra modes為候選編碼模式。由於最佳編碼模式的決定與運動估測之motion cost亦有很高的關聯性，因此第二部分依據motion cost之統計分析，找出決定候選模式集合之臨界值與QP之間呈指數函數的關聯性，加速時針對不同QP的臨界值，可直接從指數函數中求得，並將motion cost與臨界值比較以減少候選模式的選擇。第三部分為依運動向量差(motion vector difference, MVD)的動量資訊將待編碼區塊分為三種不同區域，藉由區塊特性進而減少Inter modes候選編碼模式的選擇。由修改JMVC 6.0編碼器之實驗結果顯示，本論文所提出之快速演算法平均可節省約70.79%，位元率平均下降約0.18%且PSNR平均下降約0.05dB。而與目前現有之多視角視訊之快速模式演算法相比，本論文能節省較多的編碼時間，且對RD performance幾乎不會造成影響。 The high computational complexity of multi-view video codec makes it necessary to speed up for their realization in consumer electronics. In order to adapt a fast encoding algorithm to different quantization parameters (QP), we propose a fast mode decision algorithm using motion statistics for multi-view video coding. The fast algorithm is composed of two parts. First of all, whether direct mode will be determined as an optimal mode or not based on the statistical analysis of RD cost since an optimal mode has high correlation with RD cost. The thresholds can be derived from the exponential function having high correlation with QP and RD cost. Secondly, mode candidates are reduced based on the statistical analysis of motion cost since an optimal mode also has high correlation with motion cost. The thresholds can be derived from the exponential function having high correlation with QP and motion statistics. Finally, a current MB is classified into three different kinds of regions according to analysis of motion vector difference (MVD), and the candidates for mode decision are derived from the corresponding region. Our experimental results show that the proposed scheme reduces up to 70.79% of encoding time with only 0.05 dB loss in peak-to-noise ratio (PSNR) and 0.18% bitrate decrement compared with the original JMVC 6.0. Compared with other algorithms, the proposed algorithm can reduce more computational complexity with negligible degradation of coding efficiency.
