隨選視訊服務視訊資料之解析及瀏覽

Abstract

在隨選視訊服務中，視訊資料是主要的資料元件。由於其資料量非常 龐大，而且視訊內容之認定相當主觀，見仁見智，很難客觀地以電腦來描 述。因此為了讓使用者能在大量的視訊中迅速找到所要的視訊，我們需要 一個有效地儲存和管理大量視訊的資料庫系統。 在這篇論文中，我們研發隨選視訊服務中視訊資料庫的核心技術，並 應用於新聞影片資料庫。此核心技術可以很容易的轉移到各種隨選視訊服 務上。在我們的發展環境中，採用主從架構的模式，提供使用者透過網路 查詢，並由視訊伺服器提供隨選視訊的服務。在我們視訊資料庫系統中， 使用者可以用視訊的屬性、關鍵字、及中文內容敘述做查詢。我們採用一 些適當有效的視訊解析方法和視訊瀏覽技術來幫助使用者更快找到所要的 視訊。由於視訊以MPEG-II格式壓縮儲存，我們可以藉由變動向量來偵測 場景變化，即時擷取出關鍵畫面，並做分群以組合出視訊海報，不必事先 做解壓縮。如此不但節省儲存空間亦節省處理時間。而視訊海報和關鍵畫 面的階層式瀏覽即可提供使用者友善的視訊導覽。我們亦提供使用者可以 使用瀏覽器查詢，而不受限於作業系統的平台。 In Video on Demand (VOD) services, videos are the major information elements to be processed. The volume of video is very large and it is difficult to describe and to model the content of the video. To provide users fast retrieval of desired video from a massive repository of videos, there must be a video database management system to store and manage the videos. In this thesis, we develop a prototype of video database system for VOD services. Although our system is designed for News on Demand service, the core technologies of this news video database system can be applied to other VOD services easily. In our developed environment, the system is constructed using client-server model, and supports continuous playback of videos by Video Server Engine. In the video database system, users can query by features, keywords and description in Chinese about video content. Some efficient parsing methods and browsing techniques are adopted to help users speed up the access. Video data are compressed in MPEG-II format. In the preprocessing of video data, scene changes are detected in real time using motion vector without decompression of video data, and key-frames are extracted and clustered to assemble into video posters. So, these techniques save storage space and processing time. The video posters and hierarchical browsing of key-frames are provided for friendly navigation. Additionally, in this system, users can query in WWW Browser without the limitation of OS platform.