近年來，隨著寬頻網路時代的到來，在網路上傳送多媒體服務已相當的普及，但多媒體視訊資料經過壓縮編碼之後，會產生不同重要性之視訊封包，若一視同仁的傳輸於網路上而遇到網路壅塞時，將使得視訊品質遭受到嚴重的影響。 基於以上的問題，本論文首先針對不同畫面複雜度的視訊錯誤蔓延特性進行分析，並提出一個同時參考時間域及空間域的視訊封包重要性分類機制(SC-TS)，本機制除了從時間軸的播放順序來判斷封包的重要性之外，亦同時利用錯誤追蹤(error tracking)的觀念，來決定目前畫面的每個封包被下一張畫面像素(pixel)參考的比例，並進而區分出同一張畫面的每個封包不同的重要性。為了讓本機制可以適用於各種不同特性的視訊序列，本論文進一步提出在空間域和時間域之調適性視訊封包重要性分類機制(ASC-TS)，以GOP為基礎並啟動錯誤蔓延比例的學習機制，把目前GOP的錯誤蔓延比例作為下一個GOP決定封包重要性分類的參考依據。 模擬結果顯示，相對於傳統以畫面位置為基礎的分類機制，在相同的網路環境下，本論文提出的機制能改善接收的視訊品質達0.7dB。 In recent years, the delivery of video streaming services in Internet is popular and full of potential. However, an equal error protection scheme to all video packets in Internet will significantly degrade the video quality since the encoded and packetizated video packets have different significances. Therefore, this thesis proposes a Significance Classification mechanism (SC-TS) to classify the video packet importance from Temporal and Spatial domain simultaneously. SC-TS not only determines packet significance from the frame order in time domain but also utilizes error tracking concept to differentiate significances of video packets in the same frame. Moreover, for satisfying various video sequences with different coding properties, this thesis adds a learning algorithm of error propagation property to SC-TS, which is named Adaptive SC-TS (ASC-TS) in this thesis. While utilizing ASC-TS, the required error propagation ratio of next GOP is learned from the error propagation results of current GOP. Simulation results reveal that the proposed ASC-TS mechanism can effectively improve the received picture quality up to 0.7dB under the same network environment, compared with the traditional classification mechanism that determines the packet signification from temporal domain only.
