基於FMO及收斂性動態預測的不對等保護方法

Abstract

在這篇論文裡，我們提出一個在多錯誤的網路環境下傳輸影像的UEP (Unequal Error Protection) 系統。我們使用含在H.264標準裡的工具 ─ FMO (Flexible Macroblock Ordering)，並提出了失真預測方法來對所有的Macroblocks分類。為了讓錯誤隱藏更有效率，使用基於Dispersed模式所衍生出的新模式搭配K-mean分群演算法來分配Macroblock到各個Slice groups。最後，我們提出收斂性動態預測來進一步加強所提出的UEP系統。實驗結果顯示我們的方法的確完善地保護重要的資料，且收斂性動態預測更加強了效果。

In this thesis, we present an unequal error protection (UEP) scheme for video transmission over error-prone networks. The Flexible Macroblock Ordering (FMO), included in H.264 standard, is used to accommodate different slice groups with different importance in a frame. We propose a distortion estimation scheme to define the importance of macroblocks and then assign macroblocks to slice groups based on a variation of the dispersed FMO mode and k-means clustering algorithm. Finally, a Converged Motion Estimation (CME) is proposed to further improve our proposed UEP scheme. The idea behind the proposed CME is to make important macroblocks concentrated on few only, so that the use of redundancy for error protection will be more efficient. The experimental results show that the important data is well protected using our proposed UEP scheme, and CME further improves the performance.