對於提升解碼效率一個有效的H.264/AVC NAL Unit封裝方式

Abstract

線上觀賞即時影音在過去幾十年已經成為主流的影片觀賞方式。多媒體影音為了以較小資料量在網路上傳播，通常會先壓縮再傳輸，而視訊壓縮後的資料單位稱為NAL Unit (Network Abstraction Layer Unit)。由於源自多媒體影音，NAL Unit資料有先後順序的特性，因此與其搭配的協議經常為RTP (Real-time Transport Protocol)。本篇論文提出多個能提升解碼效率的NAL Unit封裝方法。 NAL Unit的封裝方法有許多種。這些方式中，一個NAL Unit封裝入一個RTP封包的方式是最簡單且最為普遍的封裝方式。但是，由於視訊壓縮每個畫面會互相參考的特性，NAL Unit之間也存在相依性。當視訊經由錯誤更正碼處理並在有雜訊的網路環境中傳送時，一個NAL Unit封裝入一個RTP的方式可能會因為沒有考慮NAL Unit間的相依性而造成低落的解碼效率。在提出的封裝方式中，LT Codes (Luby Transform Codes) 是所使用的FEC錯誤更正碼，且封裝RTP封包時有考慮到NAL Unit的相依性。實驗結果顯示考慮NAL Unit相依性所提出的封裝方式，能夠改善解碼的表現。

Watching on-line video has become more and more popular in the past decades. Usually, Multimedia is compressed before we transmit it. And NAL Unit (Network Abstraction Layer Unit) is the data unit after video compression. Considering that NAL Unit is derived from Multimedia, ordering is its characteristic. And it’s common to see it cooperate with RTP (Real-time Transport Protocol). In this thesis, several encapsulation methods are proposed, which effectively encapsulate NAL Units into packets to achieve better decoding performance in terms of PSNR. There are many ways to encapsulate NAL Units into packets. Among them, single NAL Unit Packet encapsulation, which means that a single NAL Unit is encapsulated in a single RTP packet, is the simplest one and is widely used. However, due to frame dependency which is resulted from inter-frame prediction in video encoding, there is also dependency among NAL Units. Without considering dependency among NAL Units, the single NAL Unit Packet encapsulation scheme may result in poor decoding performance when the video is delivered with FEC and over error prone networks. In the proposed methods, LT Codes (Luby Transform Codes) are used in FEC layer and NAL dependency is considered in RTP encapsulation. The experimental results demonstrate that by considering dependency among NAL Units, the proposed encapsulation method can improve the decoding performance.