Title: 運用語意鏈結來對視訊資料庫作查詢
Video Database Query Based upon Semantic Link
Authors: 梁文嘉
Liang, Wen-Chia
羅濟群
Lo Chi-Chun
資訊管理研究所
Keywords: 語意鏈結;視訊資料庫;查詢;內容檢索;semantic link;video database;query;content-based retrieval
Issue Date: 1995
Abstract: 由於視訊資料包含豐富的資訊量,資料的內容查詢將是一項視訊資料庫運 用的關鍵成敗因素。在一般的視訊索引系統中,建立了大量的視訊資料索 引值,再以此對視訊資料的內容作查詢,為一經常使用的方式。在現今視 訊資料的物體辨識率不高的情況下,減少使用者必須建立視訊索引值的數 量, 以達成使用者能快速建立視訊資料索引值,是本研究的目的。在本 篇論文中,為了有效的減少使用者建立視訊索引值的數量,採取以下的步 驟:一、本研究在許多的視訊資料中,找出視訊資料的索引值,索引值透 過屬性與值的制訂,組合為『視訊物件』,透過視訊物件的建立與操作, 使用者可對視訊資料的內容作查詢。二、本研究透過重複出現的視訊物件 ,萃取出系統所需要的『通用視訊物件』及與主題有密切相關的『主題物 件』,其他的物件則稱為『子項物件』。三、現有的視訊資料庫系統的視 訊物件因為透過編號的處理,常會喪失原有視訊物件語意上的關聯性,因 此有必要將物件的『語意鏈結』列入系統的設計考量,本研究將子項物件 與通用物件、主題物件做出鏈結,物件語意的鏈結運用確定因子法,將物 件的語意作兩兩之間的鏈結。四、依據同類物件共同存在於同一個視訊資 料檔的特性,本研究利用貝氏機率的原理,以『物件共生機率』衡量『類 似型態物件』共同存在於同一視訊資料檔的可能性。在使用者輸入查詢條 件後,若查詢的物件為有建索引值的通用物件或主題物件,可採用直接查 詢,否則採用間接查詢的方法。間接查詢的方法,可透過語意鏈結的設計 ,以語意相關聯的物件,運用機率推理樹找出使用者要找的視訊資料。最 後,本研究利用前述所提出的方法實作此一視訊資料庫查詢系統,而從系 統執行的結果,可以明確的發現,透過本研究的設計方式,利用自動化索 引的技術,找出通用物件與主題物件,使用者只需建立少量的子項物件, 並利用語意鏈結的方式,即可達成比一般系統更好的查詢效果。 With the huge volume of data of the video database, data retrieval is the key to the success of using the video database. In most existing video indexing systems, user must build a large number of indecies for the purpose of querying the video database. How to reduce the number of video indecies is the major goal of this thesis. In order to effectively reduce the number of video indecies, the following steps are taken: Step 1, video indecies are extracted from the video data. Here, index, also called video object, is composed of attributes and their associated values. Step 2, we identify those video objects usually found in video data, and define them as "general object". For those general objects having special meaning, we call them "theme object". For objects which are neither general object nor theme object, we call them "sub-item object". Step 3, we use some "semantic links" to help we to build the semantic link of objects, for the lack of semantic of coded object identifier. In thesis , using certainty factor method to links the semantic of object. Step 4, We use Baysian probability to build "object cross-live probability table" , the relationship about two general object is living in one video segment. As to the query processing, the query object has to be input first. If the object has been indexed , like general object or theme object, the "direct query" is used, otherwise the "indirect query" is used. Indirect query use semantic links to find the object with semantic relationship. then we can use probability reasoning tree to find video data. Finally, using above method, we build a video database query system. From computational results of the prototype system, we can get the same query effect by using fewer video objects and semantic link mechanism.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT840396008
http://hdl.handle.net/11536/60539
Appears in Collections:Thesis