近年來，多視角視訊應用興起，提供了觀賞者新的視覺感受，因此多視角視訊壓縮技術成為傳輸與儲存多視角視訊的關鍵角色。相對於傳統視訊壓縮技術開發時間方向的資料多餘性，多視角視訊編碼器還可開發視角間方向的資料多餘性，但為了獲得最佳的編碼決策，龐大的運算量使得多視角視訊編碼器不易應用於即時系統。因此，本論文提出快速演算法加速多視角視訊編碼器，演算法分成兩個部分，快速畫面間預測方向決策及以epipolar geometry為基礎之視角間預測。快速畫面間預測方向決策藉由預測畫面內容編碼區塊之運動向量特性，快速決定區塊可能採用的預測方向。此外，本論文改良現有的epipolar geometry為基礎之視角間預測演算法，使用全域搜尋樣板，擴大視差向量搜尋範圍，提高視差向量估測精確度，並降低預測殘餘資訊，快速畫面間預測方向決策節省15%~24%的編碼時間，結合了本論文提出之Epipolar Geometry為基礎之視角間預測，當PSNR下降可接受情況下，整體編碼時間節省20%~28%，且獲得位元率下降0.6%~8.0%。 In recent years, various applications of multi-view video provide viewers a new viewing experience. Accordingly, multi-view video coding (MVC) plays a key role in transmission and storage of multi-view videos. However, besides temporal prediction which is used to reduce temporal redundancy, the MVC encoder also explores the inter-view redundancy by inter-view prediction. Because of the optimal decision of inter-view prediction and temporal prediction, the heavy computational load results in the difficulty of its realization in real-time systems. Therefore, this paper proposes a fast algorithm for the MVC encoder. There are two parts of our proposed scheme, fast inter frame direction predictor and epipolar geometry based inter-view prediction. The fast inter frame direction predictor decides the prediction direction that the current block may prefer, according to the motion characteristic of the current block inferred from the blocks in the neighboring frames. The epipolar geometry based inter-view prediction improves the accuracy of the location of matching blocks. Our experimental results show that the fast inter frame direction predictor reduces 15% to 24% of encoding time. By combining it with epipolar geometry based inter-view prediction, there is 20% to 28% of encoding time is reduced with the bitrate reduction ranging from 0.6% to 8.0%.
