可調視訊編碼多層編碼控制的快速演算法理論的位元與失真關係函數初步分析

Abstract

以可調整視訊編碼(SVC)之架構為基礎，本論文闡釋一個多層編碼控制下的快速演算法。由於傳統由下至上編碼控制的增進層所利用的基層資訊，其適用性不一，相較於單一層編碼會有不對稱的編碼效能損失，而多層編碼控制能在基層和增進層之間的編碼效能做權衡，此編碼方法在先前就被提出。而我們先前的研究，修正了多層編碼控制的公式，使編碼效能更具有可預測性。由於多層式編碼控制的編碼速度過於緩慢，本論文沿用之前的公式，利用時間性資訊和隨編碼內容更新的特性，提出了一個兩段式的快速演算法。根據實驗結果，本快速演算法和徹底式搜尋的比較之下，減少了平均71~79%的時間，且在編碼品質上無太多失真。此外，實驗結果也顯示，本論文提出的快速決策演算法和過去的快速決策演算法相比較，更接近徹底式搜尋的編碼效能，顯得更具有穩定性。 另外，鑑於過去的快速決策演算法通常以統計學為基礎進行分析，缺乏理論依據。我們以H.264/AVC和SVC為基礎，提出了位元配置和失真之間的關係公式，並對Mode Decision的影響因素進行理論分析。期待未來可以應用到多層編碼控制。本研究目前還在探索階段。

In this thesis,a fast mode decision for SVC multi-loop encoder control is proposed based on statistics analysis. In conventional bottom-up encoder control,enhancement layer(EL) coding is characterized by its uneven distribution of rate-distortion loss relative to single-layer coding. for trading off the coding efficiency of the base layer and the enhancement layer,a multi-layer encoder control method was proposed. In our previous study,a modified scheme of multi-layer encoder control is presented to provide a new decision criterion which are more predictable and consistent with different choices of the weighting factor. Continuing from previous study,we proposed an two-stage and context-adaptive algorithm with the help of temporal information. Experimental results show that proposed fast mode decision algorithm can save approximately 71~79% of total encoding time compared to conventional multi-layer encoder control,with only little degradation in the rate-distortion performance. The results also show that the proposed algorithm is more stable and,because the RD- performance is closer to exhaustive search. Described below and above are independent,n this thesis,we also study the theoretical analysis of mode decision. Previous fast mode decision is usually based on statistics analysis rather than theoretical analysis. We study the theoretical analysis of mode decision for H.264/AVC and SVC,propose a formulation between bit allocation and distortion,and add the factors of mode decision .Look forward to extending to multi-layer encoder control. This research is in progress