指導教授：丁建均臺灣大學：電信工程學研究所吳泊泓Wu, Po-HungPo-HungWu2014-11-302018-07-052014-11-302018-07-052013http://ntur.lib.ntu.edu.tw//handle/246246/264265In my dissertation, there are two main applications of computer vision. The first one is salient region detection improved by PCA and boundary information, and the second one is banknote reconstruction from fragments by image registration and convex quadratic programming. Salient region detection is useful for several image-processing applications, such as adaptive compression, object recognition, image retrieval, filter design, and image retargeting. In this dissertation, we propose a novel method to determine the salient regions in images. The L0 smoothing filter and a Principal Component Analysis (PCA) play important roles in our framework. The L0 filter is greatly helpful in characterizing fundamental image constituents, i.e., salient edges, and for simultaneously diminishing insignificant details. Therefore, we can derive more accurate boundary information for background merging and boundary scoring. A PCA can reduce the computational complexity, as well as attenuate noises and translation errors. A local-global contrast is then used to calculate the distinctiveness. Finally, we take advantage of image segmentation to achieve full-resolution saliency maps. Our proposed method is compared with other state-of-the-art saliency detection methods, and is shown to yield higher precision-recall rates and F-measures. Due to a variety of accidents, banknotes may be broken into several fragments. These fragments are usually stained, burned, partially lost, and twisted, which makes banknote reconstruction a hard problem. Since the fragments are always not intact, the traditional edge and texture based fragment assembling methods cannot be applied here. In this dissertation, we develop a framework for banknote reconstruction using registration and optimization. We applied the image registration using the SIFT and RANSAC. Moreover, convex quadratic optimization based on maximizing the reconstructed area and avoiding overlapping is adopted. Simulations are given to demonstrate the effectiveness of our framework.口試委員會審定書 # 誌謝 i 中文摘要 ii ABSTRACT iii CONTENTS v Chapter 1 Introduction 1 1.1 Motivation 1 1.2 Main Contribution 1 1.3 Organization 2 Chapter 2 Salient Region Detection 3 2.1 Introduction and Related Works 3 2.2 Background 7 2.3 Framework 11 2.4 Approach 12 2.5 Simulations 22 2.6 Conclusion 35 Chapter 3 Survey of Image Registration 36 3.1 Image Registration Methodology 36 3.2 Feature Detection 39 3.3 Feature Detection 41 3.4 Transform Model Estimation 47 3.5 Image Transform 50 3.6 Evaluation of Image Registration Accuracy 51 Chapter 4 Introduction to Scale-Invariant-Feature Transform 53 4.1 Scale-Space Extrema Detection 53 4.2 Accurate Keypoint Localization 55 4.3 Orientation Assignment 57 4.4 Local Image Descriptor 58 4.5 Keypoint Matching 59 Chapter 5 Brief Introduction to Convex Quadratic Program 61 5.1 Optimization Problems 61 5.2 Convex Optimization Problems 62 5.3 Quadratic Optimization Problems 64 Chapter 6 Banknote Reconstruction 66 6.1 Introduction 66 6.2 Background 70 6.3 Framework 74 6.4 Proposed Refinements of Image Registration 76 6.5 Proposed Reconstruction Algorithm 80 6.6 Simulations 83 6.7 Conclusion 89 Chapter 7 Conclusion and Future Work 90 REFERENCE 913056941 bytesapplication/pdf論文公開時間：2014/07/15論文使用權限：同意有償授權(權利金給回饋學校)電腦視覺影像套合顯著區域偵測最佳化thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/264265/1/ntu-102-F98942124-1.pdf