指導教授：陳中平臺灣大學：電子工程學研究所詹霖Jamleh, Hani Ousamah MoradHani Ousamah MoradJamleh2014-11-302018-07-102014-11-302018-07-102014http://ntur.lib.ntu.edu.tw//handle/246246/263976The aim of this research is addressing both the influence of the limited aperture size of the optical imaging system of the camera, and the defocus aberration influence on output images in order to measure useful information such as defocus and depth through the MTF (Modulation Transfer Function), further we analyze the existing defocus levels by measuring the size of blur kernels. One of the goals of our study is to make shallow depth photos with blurry background; photographers need to use cameras such as SLR (single-lens reflex) not only for carefully choosing the best position with respect to the object but also changing the lens effective focal length or aperture size in order to obtain an artistic effect mostly desired in many types of photographs (e.g. portraits), which is not available for normal camera users who prefer to use low cost compact point-and-shot cameras; for their ease of use and convenience. Nowadays, the size of TFT-LCDs (thin-film-transistor liquid-crystal displays) is getting larger, as a result; it becomes harder to inspect defects that may exist which usually require a human visual examiner to judge the severity of the defects on the final product. These defects; so called mura (Japanese shorthand) are defined as visual blemish with non-uniform shapes and boundaries. It is becoming a very serious unpleasant effect which needs to be detected and inspected in order to characterize the LCD’s quality. Through this research, we essentially propose two contributions. One that given only two images taken under different camera parameters, we measure a reliable defocus map based on scale-space analysis, then we propagate the defocus measures over edges to the entire image using matting process, eventually we will have a refined dense defocus map, which is utilized in applications such as amplifying the existing blurriness yielding a shallow depth photos from all focused images. On the other hand, it helps extracting the foreground object shape and isolating it from the background. The second contribution is experimentally detecting many types of MURA defects on LCD panels by some low-complex effective post-processing imaging techniques. Practically; we utilize the computational photography techniques to amplify defocus levels and to detect low contrast defects such as MURA. Our Computational techniques will allow the average photographers to capture more appealing photos, and the LCD manufacturers to increase their Engineer’s efficiencies and performance. We strongly proof that this study will enable cameras and automated vision systems to embed useful computation with few user interventions.中文口試委員審定書 ii 英文口試委員審定書 iii Abstract iv Acknowledgements vi Dedication vii Table of Contents viii List of Figures xii List of Tables xvii Chapter 1 Introduction 18 Chapter 2 Background and Preliminaries 22 2.1 Geometrical Imaging and Camera Model 22 2.1.1 f-number: N 25 2.2 Point Spread Function (PSF) 26 2.2.1 Wave Optics: Airy Disk PSF 26 2.2.2 Circle of Confusion: coc 29 2.2.3 Focal Gradient 31 2.2.4 Sensor Size Effect 33 2.2.5 Defocus Aberration Model 34 2.2.6 Depth of Field: The Circle of Confusion is Fixed 37 2.3 Optical Transfer Function (OTF) 40 Chapter 3 Depth Map Estimation from Defocus Blur PSF Information 43 3.1 Introduction 43 3.2 Previous and Related Work 44 3.3 Depth from Focus Process: DFF 49 3.4 Blur Estimation from Defocus Information 50 3.4.1 Sparse Defocus Estimation from One Image 54 3.4.2 Defocus Estimation from Two Images 63 3.5 Defocus Estimation Enhancement 69 3.5.1 Sparse Blur Map Post-Processing 70 3.5.2 Image Blocks neighborhood effect 72 3.5.3 Image Zoom Calibration 73 3.5.4 Scale-Space Image Processing 75 3.6 Defocus Propagation and Interpolation 81 3.6.1 Defocus Propagation by Alpha-Matting 81 3.7 Depth from Defocus Process Implementation Using Two Images 84 3.7.1 Input Image Preparation and Smoothing 85 3.7.2 Parsavel’s Theorem (Energy Theorem) and Laplacian Filter 85 3.7.3 Calibration Process 87 3.7.4 Depth Map Measurement 88 3.7.5 Implementation Flow-chart 90 3.8 Experimental Results 94 3.8.1 Computer Environment 94 3.8.2 Cameras and Settings 94 3.8.3 Defocus Map Generation from a Single Image 95 3.8.4 Estimated Defocus Map by Two Images 97 3.9 Summary 112 Chapter 4 Automatic MURA Defect Detection and Inspection in LCD Panels 114 4.1 Introduction 114 4.2 Previous and Related Work 116 4.3 System Architecture and Approach 117 4.3.1 Pseudo-Mura Patterns 119 4.4 Mura Detection Algorithm by Segmentation 121 4.4.1 Preprocessing and Residual Image Extraction 123 4.4.2 Averaging Filter 124 4.4.3 Gradient Operation and Derivatives 125 4.4.4 The Second Derivative (Laplacian) of the Sample Image 126 4.4.5 The Fusion Operation of Two Responses 127 4.4.6 Thresholding 130 4.4.7 Morphological Post-Processing Operation 131 4.5 Experimental Results 131 4.6 Discussions 132 4.7 Summary 132 Chapter 5 Defocus Amplification and Focused Object Extraction 134 5.1 Introduction 134 5.2 Related Works 135 5.3 Image Defocus Amplification 136 5.3.1 Defocus Map Amplification Experimental Results 137 5.4 Focused Object Extraction 142 5.5 Summary 143 Chapter 6 Conclusion and Future Work 144 6.1 Summary 144 6.2 Future Work 145 Bibliography 1474681644 bytesapplication/pdf論文使用權限：不同意授權Defocus Map, Depth of Field, Mura Defect, Computational Photography, Digital Image Processing, Defocus Amplification, Depth From Defocus, Shape From Focus, TFT-LCD, Camera, Computer Vision, Shallow Focus, Defocus Map, Depth Map.thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/263976/1/ntu-103-D96943043-1.pdf