HEVC(High Efficiency Video Coding)為目前最新一代的視訊編碼標準。為了能在相同主觀影像品質下,比前一代標準(H.264/AVC)具有更佳的壓縮效率,HEVC在編解碼時引入許多新穎的架構與技術,如由編碼單元(Coding Unit,CU)、預測單元(Prediction Unit,PU)和轉換單元(Transform Unit,TU)組成的遞迴式基礎架構,與區塊合併(Block Merging)、自適應式樣點補償(Sample Adaptive Offset,SAO)和離散正弦轉換(Discrete Sine Transform,DST)等技術。雖然能達到更好的壓縮效能與影像品質,但卻大幅提高其編碼器的複雜度,使其編碼時間比前一代標準高上數倍。因畫面內與畫面間預測需大量運算各組態之位元失真成本(Rate-Distortion Cost),故其複雜度是最高的部分。在HEVC之畫面內預測中,除了提出以可調式之CU、PU和TU的遞迴結構來打破早期H.264/AVC在預測區塊是固定尺寸的限制外,預測模式上的數量也擴增為H.264/AVC的3.9倍之多。 為了減少在CU深度與PU預測模式決策時所需龐大的運算量,本論文提出一自適應式演算法以早期決策最佳CU區塊大小,同時亦能減少預測模式數量。經實驗證明後,本論文所提出之演算法能在影像品質與編碼效能上取得良好的平衡。
HEVC (High Efficiency Video Coding) has been released as the latest video coding standard. In order to achieve much higher compression efficiency than that of previously published standards (H.264/AVC) under the same Peak signal-to-noise ratio, HEVC codec applies more innovative structure and architecture in the coding procedure, such as, a recursive infrastructure constituted by Coding Unit (CU), Prediction Unit (PU) and Transform Unit (TU), the Block Merging Process, Sample Adaptive Offset (SAO), Discrete Sine Transform (DST) and other technologies. While a better performance than that of H.264/AVC can be obtained, the HEVC comes with the cost of highly increased computational complexity and coding times.Among all the processes, the determination of an optimal intra- and inter-prediction mode are the most complex parts in HEVC due to large amount of rate distortion cost should be calculated before an optimal coding block size and prediction mode can be determined. To perform intra-prediction in HEVC, a recursive structure of adjustable CU, PU and TU is applied to break the limitations of a fixed coding block size in H.264/AVC. In addition, the numbers of Intra-prediction modes are highly expanded in HEVC than that in H.264/AVC. To reduce the huge amount of computations that are required for the decision of CU size and PU mode when intra-prediction mode is performed, we propose in this dissertation a fast and self-adapted algorithm so thatan optimal CU size can be determined in an early manner. In addition, thenumber of intra-prediction modes that will be sent for rate-distortion evaluation process is also confined to a smaller subset so that the coding time can be accelerated. Experimental results show that a very good tradeoff between the visual quality in PSNR, bit-rate expense as well ascoding time performance can be achieved with the proposed approach.