MPEG-2 影片上的自動化鏡頭轉換偵測法

Abstract

近年來數位影片的應用及重要性與日俱增，在多媒體系統應用中，快速搜尋影片或是存取使用者所須的資訊，第一步就是需要偵測鏡頭的變換來減低處理的資料量，本論文提出一個可依照不同影片型態能自動化設定偵測換鏡的門檻值，直接讀取MPEG-2視訊影片中的兩兩相鄰的I (P) DC畫面的DC差異值，P 畫面裡編碼方式為intra-coding 的MB個數及P畫面裡具有主要位移向量的MB個數，這三個壓縮資訊的特徵來偵測發生鏡頭變換的位置，得到相當不錯的偵測結果，針對偵測瞬間式鏡頭變換的正確性約98.7%，針對偵測漸進式鏡頭變換的正確性約92.3%，可以將此方法應用在各種類型的影片上。同時為了容易研究及增加應用性，我們也開發了一個方便使用的應用系統。

Recently, importance and application of the digital video is to grow with each passing day. And digital video has also been used widely in multimedia systems. In order to search and access user-desired information quickly, the first step is to partition a digital video into groups of meaningful segments by detecting shot change frame. In this thesis, we develop an algorithm for efficiently detecting shot changes in MPEG-2 compressed video. Since the method is based on the statistics of video difference and the statistics of information in MPEG-2 compressed video, it could be applied to various kinds of digital videos. Furthermore, we provide a useful application program of shot change detection to make use conveniently.