階層式影片內容擷取與索引

Abstract

我們利用階層式架構，提出一種有效率的影片內容擷取與索引方法。經過視訊切割，視訊群集與視訊叢聚等方法，將影片的階層式架構建立在數位影片資料庫中。對數位影片資料庫而言，影片內容擷取提供了瀏覽的能力而影片索引則提供更有效率的查詢和擷取的能力。 在我們開發的影片內容擷取與索引系統中，為了讓使用者快速了解到一部影片的內容與架構，我們提供兩種瀏覽方式供使用者選擇。一種是利用關鍵影格瀏覽影片的階層式架構。另一種是擷取影片的重要片段，並把這些片段合成影片摘要供使用者觀看。為了建立影片內容擷取與索引系統，我們建構了影片的階層式架構。而為了建構影片的階層式架構，我們提出並整合了一些視訊處理的技術，包括將影片做切割的視訊切割方法，將視訊片段群集成視訊段落的視訊群集方法和將視訊段落叢聚成視訊群組的方法。這些過程都是針對壓縮過的影片做處理，且只需要對部分的影片做解壓縮。 要評估系統所建立的階層式架構是否正確，是很主觀的一件事。為了評量系統的效能，我們請求一些使用者幫忙做鑑定。我們要求使用者在看過一些完整的測試影片後，人工地建立影片的階層式架構。我們發現使用者所建的影片階層式架構與系統所建出來的相當相似。此實驗結果證明了系統的效能是令人滿意的。

An efficient representation and indexing of video content are proposed using hierarchical structure. The hierarchical structure is constructed through video segmentation, shots grouping and scenes clustering. The video content representation provides browsing capabilities for digital video databases. The video indexing supports more efficient content-based queries and retrieval capabilities for digital video databases. In our video content representation and indexing system, we supply two kinds of methods for providing an overview of videos to users. One is the hierarchical video structure with key frames, and the other is video summary. To implement the video representation and indexing system, hierarchical video structure is constructed. For constructing the hierarchical video structure, some techniques of video processing must be investigated and implemented, including segmenting the video as shots, grouping the shots into scenes and clustering the scenes. The process is designed to work on MPEG-II sequence, where only a partial decoding is required. It is somewhat subjective how good the video structures are. The way of performance testing is to ask users to evaluate the system. We have requested several users to construct video structures after viewing the test video sequences. The grouping and clustering results of our system are very close to the results of users. The results of experiments prove that the system is satisfactory.