以全文為主的動態視訊索引法

Abstract

由於視訊(video)兼具了影像與聲音的效果, 所以在日常生活中的應用 越來越廣泛. 視訊的資料量非常龐大, 而且是一種連續性的媒體, 所以非 常需要視訊索引來輔助使用者查詢. 本論文提出一個新的方法可以動態改 變視訊索引. 在多層次(multi-layered)架構下[Chua95], [Lee97b], 將 視訊索引分成shot層與scene層. 在shot層中我們提出新的動態調整shot 的運作(operations), 如在shot中加入關鍵字, 刪除shot中關鍵字, 整合 兩個shot, 切割shot, 加入新的shot和刪除shot等. 在scene層中也提出 加入新的scene, 刪除scene等動態運作. 我們也提出利用這些動態運作, 讓系統自動建立視訊索引. 另外, 我們設計了一個原始雛型系統( prototype system)來展現上述的動態運作, 與自動建立索引的功能. Video is getting more and more popular due to its richness. Because the information contained in video is pretty much, we need to establish an index for video to facilitate the query process. In this thesis, we propose a new dynamic video indexing scheme based on a multi-layered model [Chua95],[Lee97 b]. Notable features of this thesis are to provide: (1) several dynamic video indexing operations that allow users to update video index dynamically, and (2) mechanisms to build video index automatically by using dynamic indexing operations and domain knowledge. Besides, we design a prototype system that support dynamic operations.