MPEG-2可調視訊在RSVP網路上的傳播

Abstract

在網路上傳播視訊的應用已經愈來愈重要。 這篇論文主要在探討兩 個關於視訊傳播的論題。 第一個論題討論MPEG-2可調視訊，而第二個論 題則在探討於RSVP網路上傳送可調視訊。 由於網路上的封包遺失率頗 高，因而需要可容錯的視訊編解碼方式。 其中一種方式是可調視訊。 多 層視訊編碼可提供比單層視訊更高的容錯能力。 我們專注於MPEG-2 的訊 噪比可調編碼方式。 我們也實際地寫出兩種 MPEG-2 可調的編碼器。我 們並比較了這兩種編碼方式。 除此之外，我們也改進了舊有的可調解碼 器，使之更能容忍傳輸錯誤。 我們在這篇論文提出一種在RSVP網路上 傳輸可調視訊的方法。RSVP網路提供保證模式給個別的資料流，也提供共 用頻道模式給一群資料流。 可調視訊的基礎層利用保證模式來傳送，而 加強層則利用共用模式來傳送。 我們討論兩個RSVP未提供的機制，這兩 個機制是關於資源協商和流量回饋通知。 可調的編碼方式對於控制位元 率和影像品質有較高的彈性。 在資料流較少的時候，這種方式有較高的 影像品質。並且在資料流較多的時候，可以容納更多的影像資料進入網路 。 Video distribution over the networks is becoming more andmore popular. This thesis mainly discusses two topics related to the video distribution over networks. The first topic concerns with MPEG-2 SNR scalability, whereas the second topic concerns with transmitting SNR scalable video over RSVP networks. Due to the high packet loss rate of the networks, error tolerable video coding schemes are necessary. One of the scheme is video scalability. The multi-layer coding system provides higher error tolerance than the single layer coding. We focus on the MPEG-2 SNR scalability, and implement two structures of the MPEG-2 SNR scalability encoder. Their performance are compared. In addition, an error tolerable MPEG-2 SNR scalable decoder is proposed. We then propose a transmission scheme for the MPEG-2 SNR scalable video distributing over RSVP networks. The RSVP network can provide a guarantee style for each distinct data flow, or a shared style for several data flows sharing an unreliable bandwidth. The base layer of the video is transmitted over the channel of guaranteed mode, whereas the enhancement layer is transmitted over a shared channel. We also discuss that the current RSVP is lacking in the mechanisms about resource negotiation and traffic informed feedback. The coder with scalability has more flexibility in controlling the bit rate and the image quality. With the scalability, a higher picture quality can be achieved if few numbers of streams are in the network. The scheme also allows more data streams to enter the networks.