使用平均重疊區塊補償方法應用在H.264的時間域錯誤隱藏

Abstract

在無線傳輸上時常會隨著頻帶的跳動而產生突發性的錯誤。而當這些錯誤發生時，常常會使傳送的封包產生錯誤。當視訊在播放時，這些錯誤的封包往往不但會造成失真也會影響到後續的影像持續錯誤。而錯誤隱藏是可以解決這個問題的方法。 此篇論文中，我們主要研究在H.264的時間域錯誤隱藏。我們提出三種方法並將它們整合，來回復影像中的錯誤區塊。這個整合的方法主要使用兩階層式移動向量預測搭配權重式評量方式來決定遺失區塊的移動向量，最後在使用平均重疊區塊補償的方法使錯誤區塊回復更正確。

Sometimes the burst error is happened in wireless transmission due to fluctuating channel conditions. When burst-error happens, the packets may be lost. That will cause not only the degraded quality but propagate error until an I-frame arrived successfully. To avoid these situations, error concealment at decoder is often necessary. In this thesis, we will focus on temporal error concealment in H.264. We proposed a two-level MVs prediction for choosing candidates MVs and then a weighted external boundary match algorithm as the measure criteria to determine the best MV. After the best MV is chosen, we propose a method called average overlapped block motion compensation (AOBMC) to recover the lost MB by overlapping it with multiple MBs. By combining the proposed three techniques (two-level MVs prediction, WEBMA, and AOBMC) , the experimental results show that we can recovery the lost MBs with better Peak Signal to Noise Ratio (PSNR) and visual quality.