用二維C型樹做視訊資料索引與相似擷取

Abstract

視訊資料含有大量的空間與時間資訊。為了管理多樣的視訊資料，一個可 以儲存視訊資料、維護管理視訊索引、處理視訊查詢、和擷取視訊資訊的 視訊資訊系統是很需要的。張系國博士等人曾提出一個二維串列方法，很 有效的成為影像資料的空間關係索引結構。在本論文中，我們將這種圖像 式方法應用到視訊資料索引結構上。我們研創出一種二維C型樹的知識結 構來展現視訊資料內容中空間關係的特性。這種新的表示法，不僅可以完 全保存原始影像中各物件之間所有的空間關係知識，而且更適合進行空間 知識推理與影像擷取。這項資料索引技巧可用來表示個別畫面內的空間關 係內容，並且為一個視訊資料組成一個有序列性的二維C型樹集合。這個 有序的二維C型樹集合就成為該視訊資料的代表象徵和索引；而有關視訊 資料的相似性查詢擷取問題，就演化成資料索引間的相似性比較與計算其 個別畫面間的差異變化問題。我們也發展一個雛形系統，來驗證以二維C 型樹做為視訊資料索引與相似擷取方法的效用。 Video data contains a large amount of spatial and temporal information.A video information system must be provided to manage a variety ofinformation in video data, including stoing video data, maintainingvideo indices, processing video query and retrieving video information.Chang et. al. had proposed an effective 2D string approach for spatialindexing of image data. In this dissertation, we extend this iconicapproach and apply it to video data indexing. We investigate a new knowledgemodel, 2D C-tree, to characterize the spatial information of video content.Not only does the new representation keep the whole spatial relationshipsamong objects in an original image, but the ordered labeled tree is moresuitable for spatial reasoning and image retrieval.The indexing technique represents the spatial content within individualframes and an ordering set of 2D C- trees for a video sequence is constructed.The ordering set of 2D C-trees then becomes the representative indexing of avideo sequence. The similarity retrieval of video matching problem becomesthe problem of video sequence matching by computing the similarity, or theminimum cost of matched frames. A prototype video information system is alsodeveloped to validate the effectiveness of video data indexing and similarityretrieval based on 2D C-trees.