解碼端視訊預測技術與壓縮系統設計之研究

Abstract

在傳統多媒體通訊系統中，解碼端有低運算複雜度的設計限制。隨著半導體技術進步，解碼端的運算能力越來越高。在解碼端可能採用先進的影像處理技術增加系統效能。本論文討論如何利用解碼端影像技術增進系統效能。我們將這樣的概念利用在兩個不同的應用上：分散式編碼系統和穩健編碼系統。 分散式編碼系統利用解碼端影像處理技術來作畫面間預測，系統中將複雜的畫面間預測從編碼端移到解碼端，因此可以大幅度降低編碼端的複雜度。我們提出利用解碼端畫面間預測誤差分析來最佳化編碼系統。我們提出預測誤差分類的概念，先分析解碼端預測誤差分佈然後分類，讓每一個分類裡的預測誤差統計特性接近。最後再分別將每一類誤差編碼。這樣分類編碼的方法可以節省3%~10%的位元數。我們更進一步提出基於視覺分析的編碼方法，我們提出解碼端預測誤差的視覺分析，從視訊紋理分佈、運動分佈、、、等特性分析視覺誤差的分佈，接下來加強視覺誤差的編碼。這樣的基於視覺分析的編碼方法和傳統編碼法相比可以節省10%~18%位元數，在主觀視覺上更有明顯的改善。我們提出了分散式編碼系統中編碼方法的非平等數據保護版本，實驗結果顯示我們提出的編碼法可以提供非平等數據保護的特性，這個特性可以開啟很多分散式編碼系統新的研究方向。 我們也將解碼端影像處理技術用在穩健編碼系統的設計。我們基於階層式B畫面編碼架構，提出了兩個穩健編碼系統。第一個系統是利用一種混合的階層式B畫面編碼架構可以提升解碼端錯誤修復的效果。第二個系統是基於多重描述編碼，我們利用最佳化的方法來設計一套可以根據影像內容和網路品質動態調整的編碼方法，實驗結果顯示我們提出來的系統都有較好的編碼容錯能力。

In traditional multimedia communication systems, the computation complexity of the decoder is constrained. Due to advance of semiconductor technologies, the computation capability of decoder increases. It is possible to apply advanced image processing technologies on the decoder side. This thesis discusses how to utilize decoder side image processing technologies to improve system performance. We apply this concept on two applications: distributed video coding and robust video coding. Distributed video coding adopts decoder side video prediction technologies to do inter-frames coding. The systems move complexity of inter-frame prediction from encoder to decoder, so the complexity of encoders decreases dramatically. We utilize analysis of inter-frame prediction error to optimize the compression system. The error classification is proposed: the decoder analyze macroblocks and classify them into different groups based on prediction error characteristics. Macroblocks in the same group have similar error characteristics and are encoded by a channel code. The classification based coding method would save 3%-10% bit rates. Furthermore, we proposed a perceptual based coding method. Perceptual error on the decoder side is identified by analyzing image texture distribution, motion behavior, and so on. Wyner-Ziv (WZ) bits are then concentrated on correcting these perceptual error. Compared with original coding system, the proposed perceptual based coding can save 10%-18% bit rates. Subjective quality improvement is even more. We also apply decoder side video processing technologies in robust video coding systems. Two systems are proposed based hierarchical-B coding structures. The first system adopts a hybrid hierarchical-B structure to improve the quality of error concealment method. The second system is based on multiple description video coding. We proposed a R-D optimization framework that adaptively allocates redundancy based on video content and channel conditions. Experimental results show the proposed system can provide good error resilient capability.