洪一平臺灣大學：資訊工程學研究所詹力韋Chan, Li-WeiLi-WeiChan2007-11-262018-07-052007-11-262018-07-052004http://ntur.lib.ntu.edu.tw//handle/246246/53670本篇論文的目的在於建構一個環物影片內容檢索系統，藉由此系統，使用者可以方便地從龐大的環物影片資料庫中，基於環物影片內容檢索出所需要的環物影片。環物影片相較於三維物件具有更接近真實的展示能力，同時具有容易取得及提供足夠互動的優點。為取得環物影片，可以利用相機對真實的物件從各個不同的視點取像，所擷取出的一組影像則稱為此物件的環物影片；其中所產生影像的數量可視為相機對物件取樣的密度，理論上，兩個相似的物件各自產生的環物影片，其位置上相對應的影像應該也會相當相似。因此，累加所有相對應影像的差異距離可以代表兩個環物影片間的相似度。基於以上的想法，在此論文中，我們設計了兩種特徵描述來表示一個環物影片的組成，根據這些特徵描述，我們定義了兩個環物影片之間的相似度測量值。在我們的系統中，可以用兩種形式的輸入來執行檢索，一種是用目標物件的環物影片，另ㄧ種是只以目標物件少數幾個視點所取得的影像。最後，我們提出了一套使用於環物影片檢索的關聯回饋技術。透過關聯回饋，讓我們可以更清楚的知道使用者所欲查詢的概念。The aim of this thesis is to build a content-based object movie retrieval system. It is to retrieve the desired object movies for a user from a large object movie database, based on the contents of object movies. Object movie is a good representation of a physical object because it can provide 3D interactive viewing effect, while not requiring 3D reconstruction. An object movie refers to a set of images captured from different perspectives around a physical object. An object movie can be mapped into a manifold in the feature space. Two different sets of feature descriptors, dense descriptor and condensed descriptor, are designed to sample the manifold. Based on these descriptors, we define the dissimilarity measure between the query and the target in the object movie database. The query we considered can be either a complete object movie or simply be subset of different views of the object. An algorithm for relevance feedback is also proposed in our system to capture the concepts a user has.CHAPTER 1 INTRODUCTION 1 1.1 3D Geometric Model Representation 1 1.2 Object Movie Representation 2 1.3 Content-Based Object Movie Retrieval 4 1.4 Thesis Organization 6 CHAPTER 2 RELATED WORKS 7 2.1 Content-based Approach to Object Retrieval 7 2.2 Relevance Feedback 10 CHAPTER 3 PROBLEM DEFINITION 12 3.1 Retrieving Scheme 12 3.2 Relevance Feedback 13 CHAPTER 4 FEATURE DESCRIPTOR 15 4.1 Dense Descriptor 15 4.1 Condensed Descriptor 18 CHAPTER 5 DISSIMILARITY MEASURE 20 5.1 Dissimilarity Measure 20 CHAPTER 6 RELEVANCE FEEDBACK 23 6.1 Search Process for Relevance Feedback 23 6.2 Our Proposed Approach 25 CHAPTER 7 EXPERIMENT AND RESULT 28 7.1 Object Movie Databases 28 7.1.1 First Database: OMDB1 29 7.1.2 Second Database: OMDB2 30 7.1.3 Third Database: OMDB3 34 7.2 Retrieval Without Relevance Feedbacks 36 7.2.1 Use of Object Movie Database: OMDB1 36 7.2.2 Use of Object Movie Database: OMDB2 41 7.2.3 Use of Object Movie Database: OMDB3 44 7.3 Feedback Evaluation 46 CHAPTER 8 CONCLUSION AND FUTURE WORK 49 8.1 Summary 49 8.2 Future Work 502878410 bytesapplication/pdfen-US內容檢索環物影片相似度測量值關聯回饋Relevance feedbackSimilarity measureObject movieContent-based retrievalthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53670/1/ntu-93-R91922079-1.pdf