高效能視訊編碼之適應性迴圈濾波器研究

Abstract

本論文針對高效能視訊編碼中適應性迴圈濾波器進行研究，適應性迴圈濾波器 可去除重建畫面中量化後的雜訊，提升重建畫面的品質。高效能視訊編碼中 適應性迴圈濾波器是基於整體畫面資訊做濾波器優化，必須等待整張畫面編碼完畢之後暫存所有區塊的資訊以取得最佳濾波器係數，不僅需耗費額外記憶體空間儲存，同時也造成編碼延遲的上升。直接套用先前已重建畫面的濾波器係數於目前的編碼畫面，雖解決編碼延遲的問題，但此方式並非該編碼畫面的最佳濾波器係數設計。為每個區塊設計一組目前編碼畫面中最佳濾波器係數，雖有效解決上述問題，但濾波器係數數量可能因畫面大小而呈正比成長。在不考慮編碼延遲上升的情形下，本論文探討適應性迴圈濾波器對於畫面品質的改善極限，提出基於整體畫面資訊之濾波器優化演算法，利用階層式四分樹切割方式將畫面作分割後再以Bottom-up的方式進行不規則合併。 期望達成單一區塊編碼延遲並同時取得區塊之最佳濾波器係數。實驗結果顯示本論文提出的演算法在解析度1920x1080之視訊序列下平均達0.3%的壓縮效能提升，而雙區塊編碼延遲之設計則造成0.3%的壓縮效能下降，但兩種方法皆能有效地提升重建畫面的主觀品質。

The adaptive loop filter (ALF) is mainly used to reduce the quantization noise of reconstructed pictures, and provides, at the same time, a better quality of reference frames for coding future frames. In the High Efficiency Video Coding (HEVC), an optimal filter is derived by minimizing the square error between original and reconstructed frames. Such operation, however, needs an extra memory buffer to store the information, such as all the original and reconstructed pixels, of the current frame, and thus the encoding latency is inevitably risen up. To alleviate such problem, a so-called time-delay filter, which is an optimal filter optimized for a previously coded frame, is applied as if it were found for the current frame. To keep the encoding at single block latency while retaining the filters to be optimal for the current frame, this thesis studies the trade-offs between the coding performance and the encoding latency. A quadtree-like filter optimization algorithm is proposed to improve the coding performance at the cost of a higher rate overhead for signaling filters. As compared with the HM-6.0 software, the proposed scheme achieves BD-rate saving of 0.3%. A negligible BD-rate loss of 0.3% is observed for the two-blocks encoding latency scheme. Although the two approaches do not bring any impact on the coding performance in terms of BD-rate saving, they can still provide noticeable improvement to the visual quality of the reconstructed frames.