使用鄰近邊界匹配之時間性錯誤隱藏演算法

Abstract

錯誤隱藏是個能有效減少解碼端發生錯誤時所造成影響的一種技術。當解碼端在解碼時若發生錯誤，則解出來的圖片可能會產生令人不愉快失真現象。在本論文中，吾人提出了二個對於H.264的時間性錯誤隱藏演算法。 第一個方法為多重邊界匹配（Multi-sides Boundary Matching, MSBM）演算法，先將一個block切割為四個8×8的子區塊，再利用與損壞區塊最接近的二個邊的邊界資訊來補回來。另一個方法為鄰近邊界匹配演算法，利用受損block的四周四個區塊及共位區塊的(Collocated Block)的移動向量經過排序後取中值，將其所指到的區塊在一小的搜尋範圍中找到最佳的候選區塊。 實驗結果顯示吾人所提出的兩種演算法比遞回性區塊耦合(Recursive Boundary Matching, RBM)方法有較佳的補償效果。

Error concealment is an effective approach to reduce the influence of errors that occurs at the decoders. When error occurs, it may degrade the reconstructed pictures and lead to undesirable visual distortion. In this thesis, we propose two novel temporal error concealment methods for H.264 video streaming. The first method is a multi-side boundary matching (MSBM) algorithm. In which method, a corrupted macroblock (MB) is divided into four sub-blocks and two side information of each sub-block is used to find the best block to recover it. The second method is a neighboring side-matching algorithm, which uses the median value of the motion information of neighboring blocks of a corrupted MB and the motion information of its collocated block in the previous frame as the initial location of the best block. Then this method searches the best block in a small region. Experimental results show that the proposed two methods perform better than the RBM method.