吳家麟臺灣大學：資訊工程學研究所翁仲毅Weng, Chung-YiChung-YiWeng2007-11-262018-07-052007-11-262018-07-052007http://ntur.lib.ntu.edu.tw//handle/246246/53986在這篇論文中，我們提出一個嶄新的方法進行自動化電影分析。有別於傳統利用低階特徵值的分析概念，這個方法試圖從角色彼此之間的關係切入，提供一個電影分析的新觀點。在電影中，角色之間的互動就如同人們在社會中的行為，而電影也就如同一個小社會。我們引進了社會網路分析的概念，將電影之中角色彼此之間的互動形塑成一個網路─「角色網」。 經由分析角色網，我們提出的方法可以有效率的抽取出隱含在電影之中的關係結構，包含主角的辨認和社群的切割。同時，我們也設計了一個完整的架構來實現這個分析概念。實驗結果證實了方法和模組的有效性。此外，以角色網為基礎，將可以從不同的視野達成傳統方法所難以蹴及的語意分析。我們相信這個概念將可以為自動電影分析領域帶來新的刺激和觀點。We present a brave new way to analyze movie content, from the perspective of the relationships between roles rather than low-level audiovisual features.Interactions between roles in a movie resemble human behaviors in a society. Roles’ actions lead the story and make viewers understand what directors wantto present. In this thesis, we introduce the idea of social network analysis tomodel the relationships of actors/actresses as a network, called RoleNet. Through analyzing this network, the proposed approach automatically determines the leading roles and the communities embedded in movies. We also describe an implementation framework to realize the proposed model. Based on RoleNet, we provide some insights to facilitate more semantic extensions, which go beyond the capabilities of pure feature-based methods. Its believed that the proposed idea establishes a new way to perform movie content analysis.I RoleNet Model 1 1 INTRODUCTION 2 1.1 Motivation 2 1.2 Related Works 3 1.3 New Perspective: Social Network Analysis 4 1.4 Summary of Contributions 6 1.5 Thesis Organization 7 2 RoleNet 8 2.1 Definition of RoleNet 8 2.2 Construction of RoleNet 9 3 BILATERAL MOVIE Analysis 12 3.1 Introduction 12 3.2 Leading Role Determination 14 3.3 Community Identification 15 4 GENERALIZATION 18 4.1 Introduction 18 4.2 Leading Role Determination 19 4.3 Community Identification 22 4.3.1 Micro-community Identification 22 4.3.2 Macro-community Identification 24 II EVALUATION 27 5 EXPERIMENTAL SETUP AND IMPLEMENTATION FRAMEWORK 28 5.1 Evaluation Data 29 5.2 Implementation Framework 29 5.2.1 Overview 29 5.2.2 Face Detection 30 5.2.3 Data Processing 30 5.2.4 Face Recognition 32 5.2.5 Accuracy of Face Recognition33 6 Results 34 6.1 Ground Truth 34 6.2 Performance of Leading Role Determination 35 6.3 Performance of Community Identification 36 6.3.1 Macro-Community Identification 36 6.3.2 Micro-Community Identification 37 7 EXTENSION, CONCLUSION & FUTURE WORK 40 7.1 Extension 40 7.2 Conclusion 43 7.3 Future Work 44 References 47 Vita 512047130 bytesapplication/pdfen-US角色網電影分析社會網路分析RoleNetmovie analysissocial network analysisthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53986/1/ntu-96-R94922025-1.pdf