利用多視角深度視訊壓縮技術可將3D視訊進行有效率的壓縮，但運算複雜度高，因此，如何達成高品質合成視角，同時有效降低編碼計算複雜度為一個重要的議題。由於色彩視訊與其深度視訊編碼具關聯性，本論文提出以高品質合成視角為導向之快速深度視訊編碼模式決策演算法，共包含二個部分，第一部分為快速模式決策方案，由於深度視訊之物件與色彩視訊之物件的運動行為十分相似，而深度視訊又具有大量平滑區域，因此本論文參考色彩視訊中同位置的最佳編碼模式，以及深度視訊畫面內上方已編碼區塊的最佳編碼模式，深度畫面之複雜度，決定候選編碼模式。由於深度視訊須使用於產生虛擬合成視角視訊，因此，本論文之第二部分決定以達成高品質合成視角視訊為目標，參考色彩影像中同位置區塊之水平方向的灰階變化，判斷是否為易造成合成視角影像失真的區域，再決定合適的RD代價函式，以選擇最佳編碼模式，提高合成視角的品質。由修改JMVC 6.0.3編碼軟體之實驗結果顯示，本論文所提出之快速演算法，平均可節省約53.08%，BDBR平均上升約1.12%，且合成視角之BDPSNR幾乎無改變。3D video data can be compressed efficiency by using multi-view video plus depth (MVD) coding technology, but the computational complexity is pretty high. In order to accelerate the encoding process, the correlation between a texture video and its depth video should be explored. Moreover, how to achieve the good synthesized viewing quality is an important issue. In this thesis, we propose a view synthesis oriented fast mode decision for depth video coding. The algorithm is composed of two parts. The first part is a fast mode decision algorithm for depth video coding, which considers that object motion in texture video and depth video is similar, and depth videos has many smooth regions. We refer to the optimal mode of the co-located MB in the texture video, and the upper encoded MB in the depth video, and the complexity of the current MB in depth video. Finally, the candidates for mode decision are derived for fast depth video coding. In the second part, to achieve high synthesized viewing quality, we check the variation of intensity along the horizontal direction in the texture video. Then, an appropriate RD cost function is selected. Our experimental results show that the proposed scheme reduces up to 53.08% of encoding time with 1.12 dB BDBR increment and almost no BDPSNR loss in virtual views compared with the original JMVC 6.0.3.
