相較於單視角視訊，多視角視訊可提供使用者更豐富的資訊，但也伴隨著可觀的資料量。單視角視訊編碼僅開發時間方向之畫面間相似性，以降低多餘的運算量，而多視角視訊編碼則開發視角方向之畫面間的關聯性。因此，如何同時開發時間與視角方向的資料多餘性，並更有效降低計算複雜度，不失影像品質是值得探討的研究課題。本論文提出之快速演算法結合兩部分，分別為快速畫面間預測方向決策(FIPD)與快速參考畫面方向決策(FRPL)。快速畫面間預測方向決策在enhancement views中，可參考時間方向相鄰畫面與相鄰視角之對應區塊的運動特性，快速決定目前區塊是否僅採取時間預測方向。而快速參考畫面方向決策則進而可應用於base view與enhancement views，以16×16區塊模式之運動或視角估測之最小rate distortion cost為依據，取得目前區塊可能較偏好之參考畫面方向與畫面間預測方向(時間或視角)，快速決策該區塊其餘不同區塊模式的參考畫面方向與畫面間預測方向，以降低搜尋次數。此提出之演算法可與其他快速模式決策演算法互相結合，以達到更高之編碼效能。由JMVC 6.0編碼器之實驗結果顯示，在PSNR品質可忽略些許降低且位元率也在可接受範圍的情況下，本論文所提出之演算法在可節省約55%至62%的編碼時間。Multi-view videos can provide users rich entertaining experiences. Since they accompany with a huge amount of data, the development of multi-view video codings is stimulated. Although multi-view video coding explores the inter-view correlation, the computational complexity is also increased. Therefore, how to develop a fast algorithm without losing image quality is an important issue of MVC. In this thesis, we propose a novel algorithm by combining the fast predictor of inter-frame prediction direction (FIPD) and the fast decision of reference picture list of motion and disparity estimation (FRPL). In enhancement views, the fast inter frame direction predictor may prefer temporal prediction according to the motion characteristics of the block to be coded. Next, the reference picture list of each inter mode candidate of the current block follows the decision of the 16×16 macroblock mode of the same block in base view and enhancement views. Therefore, the inter frame prediction of one macroblock is reduced to a single prediction direction with a single reference picture list. The experimental results show that the proposed scheme reduces up to 62% of encoding time with negligible loss of PSNR and bit-rate increase compared with the original JMVC 6.0.
