隨著科技的蓬勃發展，不管是硬體設備所能提供的畫質和解析度，或是使用者期望體驗的視覺享受都隨之上升，目前的最新標準H.265/HEVC僅支援2K的視訊，但近年來已經出現許多4K的視訊內容，在未來，4K高解析度的視訊勢必成為主流，因此JVET (Joint Video Exploration Team)，從2015年開始討論最新的視訊壓縮標準H.266/FVC (Future Video Coding)，預計在2020年正式發布為國際視訊壓縮標準。 H.266/FVC的CU採用QTBT (QuadTree plus Binary Tree)編碼架構，其不僅消除CU、PU和TU的複雜層次結構，還可以支援最小8×8到最大256×256的正方形編碼區塊，甚至還依據畫面的紋理特性提供矩形的編碼區塊，QTBT架構雖然提供更好的編碼效能，但因其複雜的編碼架構，在畫面內編碼甚至提高5.6倍的編碼時間，因此，針對畫面內編碼，發展CU之快速決策演算法是非常重要的議題。 本論文提出一個畫面內CU之快速分割決策演算法，分為兩部分探討：首先第一部份利用空間特徵提前跳過CU二分樹深度(BTDepth)之編碼，並提前判斷二分樹分割模式(BTSplitMode)；在第二部份則利用空間特徵提前跳過或提前終止CU四分樹深度(QTDepth)之編碼，整合兩部分後可節省下不必要的運算來降低編碼時間。本論文所提出之方法平均可節省22.53%的總編碼時間，而BDBR僅提升0.617%。 ;With the development of technology, the requirement of hardware equipment or user’s expectation to visual enjoyment rises gradually. Joint video exploration team (JVET) started to establish the latest video compression standard, future video coding (FVC). FVC adopts quadtree plus binary tree (QTBT) based CU structure which not only removes the complicated hierarchical structure of CU, PU and TU, but also supports square coding block and rectangular coding block based on texture of video content. Although QTBT can provide better coding performance, it even increases encoding time in intra coding dramatically. Therefore, developing a fast intra CU partition decision is necessity. In this paper, a fast CU partition decision algorithm in FVC intra coding based on spatial features is proposed. Different spatial features in pixel domain are proposed in binary tree and quadtree. Spatial features for binary tree is to early skip the encoding process of CU with binary tree depth and early determine its binary tree split mode. Spatial features for quadtree is to early split or terminate CU with quadtree depth. The proposed method compared to JEM5.0 can save 22.53% encoding time with only an increase 0.617% in BDBR.
