自動回歸模型改良之多重描述編碼

Abstract

當視訊影片透過網路傳輸時，多重描述編碼是一種常被用來降低錯誤影響的技術。在本篇論文中，我們提出了一種以自動回歸模型加強的多重描述編碼。在一般的多重描述編碼當中，編碼的效能與容錯能力是評估一個方法好壞的重要標準。在我們提出的方法中使用自動回歸模型，是為了在不減低容錯能力之下，降低冗餘資訊。我們提出的多重描述編碼是由兩個對稱的描述子所組成。一個是包含有 h.264 標準的偶數幀與奇數冗餘幀，而另一個則是奇數幀與偶數冗餘幀。奇數與偶數冗餘幀都是透過自動回歸模型產生的預測幀所進一步產生出來。透過實驗證實，我們提出的方法與由內插幀預測而產生出來的冗餘幀相比，有更好的編碼效率與容錯能力。

Multiple description video coding (MDC) [1] is one of popular solutions to reduce the detrimental effects caused by transmission over error-prone networks. In this thesis, an auto-regressive model enhanced MDC is proposed. In general MDC architecture, redundancy rate and error resilience performance are important criterion for assessment. Auto-regressive model adopted in our proposal aims at reducing the redundancy rate while keeping the error resilience performance in our proposal. The proposed MDC model comprises two symmetric descriptions. One description is composed of even frames in h.264 standard and odd residual frames; while the other is omposed of odd frames and even residual frames. Both even and odd residual frames use the prediction frames generated by auto-regressive model. The experiments show that it achieves better coding efficiency and error resilience than descriptions which residual frames are predicted from interpolated frames in packet loss networks.