視訊資料自動分類

Abstract

由於視訊資料庫的發展以及WWW的使用日趨普及，未來使用者在WWW上瀏覽 視訊資料已是必然之趨勢。因此如何將視訊資料以類似“蕃薯藤”整理網 站的方式予以分門別類，藉以協助使用者瀏覽搜尋其有興趣的視訊資料， 是重要的研究課題。但是由於人工分類不僅耗時且易受分類者主觀認知的 影響，造成分類結果與使用者認知不一致，而無法有效瀏覽搜尋相關視訊 資料。故本研究之主要目的是要探討視訊資料分類之自動化，以改進人工 分類之缺點。本研究以新聞資訊為例建立分類架構，並透過問卷調查方式 ，分析不同受訪者之間的分類結果，以驗證人們對於文件之分類是否存在 分類認知的差異。本研究進而提出視訊資料自動分類方法，以視訊內容之 關鍵字描述為基礎來進行分類。不同於文字文件之分類，視訊資料具有結 構階層之特性，因此所提出的方法是依據視訊內容在視訊階層之權重值來 分類，並且考慮分類類別之間的階層關係，使視訊分類能兼顧特定性與涵 蓋性。論文最後以兩個實例說明視訊分類方法與其結果分析。 Owing to the development of video database technology and popular use ofWorld Wide Web, there is a tendency that users will browse video data on theweb. As a result, it is a key issue to classify video data into categories like the way "Yam Web " organizes web sites, in order to help users browse andsearch video data. However, human classification is not only time- consuming butis also apt to be affected by subjective cognition of classifiers, which may produce classification result that is incompatible with user's cognition so that users cannot browse and search video data effectively. The primary goal ofthis thesis is to approach how automatic classification of video data can beachieved, so as to amend the disadvantages of human classification. First, the thesis builds a classification structure by taking news data as anexample, and uses questionnaires to survey the classification results of different examinees, in order to verify whether difference in cognition ofdocument classification does exist among different people. Secondly, the thesis proposes an approach to classify video data automaticallybased on video content. Different from the classification of text documents, the classification of video data is according to the structural weight of videocontent, which tries to capture the characteristics with respect to the structural hierarchy of video data. Furthermore, with the considerations ofhierarchical relationship among classes, the proposed technique aims to achievean appropriate balance between specificity and exhaustivity on video classification. Finally we use two examples to show how the video classification algorithm is performed, as well as to discuss the analysis result.