高效率視訊編碼 (High Efficiency Video Coding, HEVC)為當前最新的國際視訊編碼標準，相較於前一代H.264/AVC視訊編碼標準，HEVC提供了各種更具彈性的編碼方式及工具，有效提升其編碼效率，在相同主觀視覺品質下，HEVC相較於H.264/AVC大約可節省一半的位元率。不論畫面內編碼 (intra coding)或畫面間編碼 (inter coding)，HEVC所採用的四分樹架構 (quadtree structure)的編碼單位 (coding unit, CU)及轉換單位 (transform unit, TU)，以及多樣的預測單位(prediction unit, PU)，均大幅增加其編碼之運算複雜度。本論文針對畫面內編碼的CU、TU深度決策以及PU的intra模式決策，提出以簡略模式成本值 (rough mode cost, RMC)為基礎之快速編碼演算法，RMC為參考軟體原始編碼過程所產生，所提方法產生之額外運算量相當低。在CU深度決策方面，由於CU切割與不切割時RMC存在一定差異，離線編碼取得目前CU切割/不切割時RMC的值，將兩畫面特性: 畫面複雜度與畫面梯度值納入考量，以曲線擬合求得CU切割/不切割的臨界成本值的模型，快速決定當前CU是否切割；在intra模式決策方面，RMC較大的模式被選為最佳模式的機會較低，因此將RMC較大的模式從候選列表濾除，減少需要測試的模式；在TU深度決策方面，我們以最小RMC模式的TU切割狀態來預測其餘intra候選模式的TU切割方式，所提方法用以取代參考軟體內建方法，且更有效率地快速決策TU。所提之整體快速演算法與參考軟體相比，平均約可節省51%的編碼時間，而僅增加約0.69%的BD-rate。;High efficiency video coding (HEVC) is the latest international video coding standard. Comparing with H.264/AVC, HEVC provides flexible encoding schemes and tools to enhance the coding efficiency. HEVC saves half the bit rate of H.264/AVC for the same subjective video quality. The quadtree-based coding unit (CU) and transform unit (TU) structure, as well as various prediction units (PUs) of HEVC, increase encoding complexity considerably in intra coding and inter coding. This dissertation proposes a rough mode cost (RMC)-based algorithm for accelerating CU/TU depth decisions and PU mode decisions in HEVC intra coding. The computational overhead of the proposed algorithm is low because the RMC is calculated during the original intra coding process in reference software. For CU depth decisions, RMC values are used to determine CU partition because RMCs of split CUs often differ from those of nonsplit CUs. Two frame characteristics, namely frame complexity and frame gradient, are used to model offline RMCs of split and nonsplit CUs to obtain the threshold for fast CU decision. In the case of intra PU mode decisions, modes with higher RMCs have a lower probability of being the best mode. Hence, such modes are removed from the candidate list to reduce the number of test modes. In the case of TU depth decisions, the TU partition of the mode with the least RMC is used to determine the TU partitions of the remaining candidate modes. The proposed TU partitioning method is used to replace the default method in reference software, and it demonstrates superior performance to the default method. The proposed algorithm can reduce the encoding time by approximately 51% on average, with a 0.69% increase in the BD rate.
