應用於H.265/HEVC視窗視訊編碼之畫面內線型區塊複製預估模式

Abstract

基於畫面內複製預估方式，我們於本論文提出一畫面內線型區塊複製預估模式，將畫面內複製之基本區塊延伸至更小單元，以提升高效能視訊壓縮(High Efficiency Video Coding, HEVC)之視窗視訊編碼效率。此模式之運作方式如下：首先，將目前的預測區塊以水平或垂直的方式切割成數個寬度為一個像素大小的線型區塊；接著，所有線型區塊逐一於目前畫面之重建區域內進行畫面內複製。此模式所引用的概念近似用於無失真資料壓縮之字串匹配演算法。然而，相較於字串匹配演算法，此預估模式更能相容於高效能視訊壓縮之設計架構，且其具備較高的平行處理能力。於本論文中，我們將針對此模式之各種設計議題進行探討，包含以雜湊為基礎的快速區塊向量搜尋演算法、對於記憶體存取頻寬與CABAC吞吐量之影響。實驗結果顯示，當我們將畫面內線型區塊複製模式整合至視窗視訊編碼之測試平台SCM-1.0時，其可帶來2.0-7.5%之位元率下降，且僅造成10%的編碼執行時間上升與幾乎無影響的解碼時間。

In this thesis, we present an intra line copying mode, which introduces a finer-granularity block partitioning structure for the intra copying technique, to further improve the compression efficiency for HEVC Screen Content Coding. This mode first divides a prediction block horizontally or vertically into 1-pixel-wide lines and then operates intra copying line-by-line from the reconstructed pixels in the current frame. It shares some parallels with the notion of string matching widely used for lossless data compression. It, however, is more compliant to the current HEVC design and is parallel-friendly. This thesis shall detail its various design issues, including the design of a fast block vector searching algorithm based on hashing techniques, and the impact on the memory access bandwidth and CABAC throughput. Experimental results show that the proposed mode can offer 2.0-7.5% of BD-rate reductions, with only 10% increase of encoding runtime and negligible impact on decoding runtime, when integrated into the SCM-1.0 test model.