透過您的圖書館登入
IP:3.147.55.42
  • 期刊

使用AC-3特徵來進行電影資料的自動場景分類

Automatic Scene Classification of Movies Based on the AC-3 Features

摘要


隨著多媒體的應用範圍日漸擴大,多媒體資料內涵式分析成為當前的研究焦點。在各種多媒體資料之中,電影為一內含大量影音資訊的多媒體。而目前的數位電影資料的儲存與交換媒介中,目前仍以DVD作為標準主流格式。而在DVD所使用的影音規格中,又以MPEG-2視訊和AC-3音訊最為常見。由於電影資料過於龐大,要整部影片直接進行分析並不容易,因此將電影以電影的場景結構做分割,能有助於對電影資料作內涵式之分析與應用。在本篇論文中,我們提出一種利用AC-3音訊特徵來分類景的方法。我們先從原始電影資料中切割出各個獨立的鏡頭,再分離出各個鏡頭的AC-3音訊片段,進而取出各個鏡頭之AC-3特徵值作為景辨識的基礎,最後再佐以電影場景的特性來作為自動電影場景分類的依據。

並列摘要


As applications of multimedia are getting popular, content-based analysis of multimedia data has become of the most important research topics. Among various types of multimedia, movie plays an important role due to its rich semantics. DVD is the most popular storage format for digital movies. It uses MPEG-2 and AC-3 as its video and audio compression standard, respectively. Since the size of a movie is very large, understanding its cinematic structure will be very useful for analyzing its contents. In this paper, we propose a method to automatically classifying the scenes for a movie based on its AC-3 audio data. First, a movie is divided into a sequence of shots by a certain commercial software available.Then, we extract the AC-3 audio features for each shot. With these AC-3 audio features and properties of scene types, we can effectively classify unknown scenes into a set of pre-defined scene types.

延伸閱讀