精簡型主從式計算之多媒體傳輸與檢索技術

Abstract

近來科技的進步為我們的生活帶來了許多革命性的影響，網際網路上平台之間的協同合作益發重要，尤以多媒體資料的傳輸越來越受重視。此篇論文旨在探討精簡型主從式計算之相關議題與技術，以實現終端機伺服器與精簡型客戶端之間的多媒體應用。主要研究方向為：(1)協同式視訊會議系統，以及 (2)泛用編碼架構。此系統在終端機上彼此分享共用元件，同時增加了使用上的便利度。藉由伺服器端的資料預先轉碼為泛用格式而後傳送到客戶端，精簡型主從式運算環境可有效地支援多媒體應用。泛用編碼技術可將客戶端的工作量轉移到伺服器上，用以減輕客戶端的負擔。我們的實驗結果亦顯示出此架構不僅增加了客戶端的多媒體應用能力，同時降低其記憶體消耗，充分說明泛用型嵌入架構對於資源有限之精簡型客戶端其優點所在。 此外，由於許多不同形式的多媒體串流必須同步播放，時間控制上亦求嚴謹。我們在此論文中提出適應型傳輸方案，基於即時傳輸協定以確保影音串流的錯誤復原與同步播放，此適應型傳輸機制可在傳輸穩定性與減少傳輸延遲之間取得良好的平衡。 為求縮減多媒體資訊之檢索時間，我們更進一步提出了適用於高維度的最近鄰搜尋演算法。最近鄰搜尋問題意即：給定一個點集合，找出距離查詢點最近之點。此問題常見於多媒體應用層面，特別是相似度搜尋。過去所提出的解決方法其演算時間大多與維度呈指數遞增，有鑑於此，我們提出一個有效率的動態維數選擇演算法，藉此提昇最近鄰搜尋演算法在高維度的表現。

Recent technology advances have brought revolutionary impacts to our life. The Internet emerges as a collaborative platform and it is envisioned to be even more important for years to come. In particular, real-time transmission of multimedia data over the Internet is attracting an increasing amount of attention. This dissertation explores the issues and the techniques of enabling multimedia applications for the thin client/server computing. A prototype of a video communication system based on H.323 family of ITU-T recommendations is developed to demonstrate the support of multimedia applications with a terminal server and thin clients such as Windows-based terminals. Our studies on the network applications over thin clients are composed of two parts: (1) the collaborative client/server conferencing system and (2) the universal transcoder architecture. With the proposed architecture, an application of multi-user video conferencing system is devised for illustrative purposes. The collaborative client/server conferencing system leverages the benefit of sharing common components within multiple conferencing terminals as well as increases their extensibility and ease of use. As validated by our experimental results, multimedia applications can be efficiently supported in an IP-based thin client/server computing environment. In addition, the universal transcoder technique is devised to shift jobs from thin clients to a supporting server, thus reducing the computing overhead required by the clients. In addition, multimedia streams impose tight temporal constraints since different kinds of continuous multimedia streams have to be played synchronously. We devise in this dissertation an adaptive transmission scheme to ensure the stable and synchronous playback of audio and video streams based on the Real-time Transport Protocol. Realization of our adaptive transmission control is composed of a series of operations in three stages, namely, (1) dynamic reordering mechanism, (2) error-resilient mechanism, and (3) adaptive synchronization mechanism. As validated by our performance study, the adaptive transmission mechanism is able to strike a good balance of both stable playback and the end-to-end delay reduction. Furthermore, in order to shorten the retrieval time of multimedia contents, we proposed algorithm DDS for nearest neighbor search in high dimensions. The Nearest Neighbor Search problem is defined as follows: given a set P of n points, answer queries for finding the closest point in P to the query point. This problem arises in a large variety of multimedia applications, particularly in the context of similarity search. Most existing techniques for solving this problem suffer from the 'curse of dimensionality' that means the performance of system scales poorly with increased dimensionality of underlying data. In view of this, we propose an efficient dynamic dimension selection algorithm to improve the performance of the nearest neighbor search especially in high dimensions.