使用多重參考幀在基於R-D最佳化之H.264錯誤恢復影像編碼

Abstract

對於高壓縮率影像編碼，如H.264，因為潛在的錯誤傳遞效應，要透過錯誤傾向環境來進行影像傳輸是一項挑戰。在本篇論文中，我們提出了一個架構在H.264 影像編碼下，並結合Error Resilient Rate-Distortion Optimization (ER-RDO) 及多重參考幀 (Multiple Reference Frames, MRF) 的編碼策略。因為錯誤傳遞(Error Propagation) 可能造成接收端影像品質的顯著下降，因此我們加入Dominant Intra Frame (DIF) 做為在多重參考幀環境下的一個參考幀。藉由DIF，可以有效地抑制錯誤傳遞，同時也可以確保編碼效率。而藉由ER-RDO 技術，也可以讓影像編碼具有更高的適應性以符合不同傳輸環境及不同的影像內容。雖然MRF 在影像編碼的效率及錯誤傳遞的抑制上有不錯的效果，然而隨著參考幀的增加，計算上的複雜度也相對地提升，尤其是在Motion Estimation (ME) 上，因此為了有效減少計算複雜度，我們也提出了一個快速移動估計演算法(Fast Motion Estimation Algorithm)。最後的實驗結果顯示，我們提出的方法能比既有架構在MRF 下的方法提供更多的錯誤恢復效果。

Transmission for high-compressed video coding, such as H.264, over error-prone environment is quite a challenge due to the potential error propagation. In this thesis, we propose an error-resilient scheme for H.264 based on Error Resilient Rate-Distortion Optimization (ER-RDO) and Multiple Reference Frames (MRF). Since the error propagation may substantially degrade the received video quality, we include a Dominant Intra Frame (DIF) as one reference frame to alleviate error propagation. By predicting from the dominant intra frame, it can be more coding e?cient and more error resilient. Multiple reference frames technique improves the coding e?ciency and suppresses the error propagation. However, it also brings a major drawback that the computational complexity for motion estimation (ME) may increase depending on the number of reference frames. To e?ciently reduce the computational complexity of ME, we propose a fast motion estimation algorithm. The experimental results show that the proposed scheme has substantial improvement over the existing schemes in providing error resilience using MRF.