位元率-失真最佳化網路視訊串流技術─H.264可調視訊編碼於實時傳輸協定進行接收端驅動重傳

Abstract

欲達到位元率-失真最佳化的視訊傳輸，一套可靠的估測模型將扮演重要角色。此研究目的為建立一套位元率與失真的估測模型，在特定的傳輸機制下，給予封包遺失率與傳輸策略，即可計算出視訊在傳輸後的位元率與失真之期望值。此估測模型考量了在有線網際網路上以實時傳輸協定傳輸視訊時，實際上所需考慮的細節，適用的傳輸機制包含接收端驅動重傳以及接收端積極重傳，且應用於傳輸可調視訊編碼。 研究結果發現，當封包遺失率愈大，不同傳輸策略所造成的差異愈明顯。且最多重傳次數愈大時，失真與位元率的變化愈趨於平緩。 透過此模型，我們提出了最佳化傳輸策略的演算法，給予有限的頻寬之下找出最佳傳輸策略，使得傳輸後的視訊具有最小的失真期望值。

In this thesis, we aim to build a rate-distortion estimation model. The expected rate and distortion can be calculated through our estimation model with a given packet loss rate and a transmission policy under certain transmission scenario. Many details need to be tackled when streaming via RTP over the wired Internet are taken into account in our estimation model. This model is build for SVC video streaming under receiver-driven retransmission and receiver-driven aggressive retransmission. The results show that the difference of RD points between different transmission policies become larger when the packet loss rate increase. As the maximum retransmission times increase, the improvement of distortion becomes less obvious. We also propose transmission policy optimization algorithms. For a given bit rate, we can find the optimal transmission policy with minimum expected distortion through our optimization algorithms.