指導教授：莊永裕臺灣大學：資訊工程學研究所陳彥豪Chen, Yen-HaoYen-HaoChen2014-11-262018-07-052014-11-262018-07-052014http://ntur.lib.ntu.edu.tw//handle/246246/261484光場相機近年來已展現了優於傳統相機的許多優異能力,它們能提供許多傳 統相機所無法達成的功能,如:數位重對焦、改變視角以及三維場景重建等。雖然 光場 相機擁有許多的優勢,但是它們成像的解析度對於現代使用者的要求來說仍然不夠高。為了要提升成像的解析度,我們期望對光場相機的原始影像作超解析度運算,雖然已經有許多作用於單一影像的超解析度演算法研究,但是針對光場相機的超解析度演算法卻很少。 在這篇論文中,我們提出了一種應用在對焦式光場相機上的超解析度演算法,經由分析對焦式光場相機的成像過程,我們發現了經由利用原始影像的多於資訊搭配上現今頂尖的自我範例超解析度方法作光場相機的延伸,能夠提供更精確的範例。我們方法的輸入為對焦式光場相機的原始影像,而輸出為超解析度運算過後的原始影像,後面我們將會展示此方法是有效果及有效率的,並且保留了光場相機做數位影像生成時的彈性。Light-field cameras have demonstrated capabilities beyond traditional cameras. They can provide users functions such as digital refocusing, changing views, and 3D scene reconstruction. Although there are a lot of advantages for light-field cameras, the resolutions of the images they rendered are not high enough for today’s user demand. To enhance the resolution of the rendered images, we are going to do super-resolution for the light-field data. There are many single image super-resolution algorithms, but the light-field super-resolution methods are still rare. In this work, we propose a super-resolution method with the focused plenoptic camera. By analyzing the image formation process of the focused plenoptic camera, we figured out a way to leverage the redundant information of the plenoptic camera to further extend the state-of-the-art example-based single image super-resolution method proposed by Freedman and Fattal to light-field. The input of our method is the raw images captures by the plenoptic camera, and the output is the super-resolved raw images. We will show that the proposed method is effective and efficient, and also retains flexibilities for plenoptic rendering.誌謝 i 摘要 ii Abstract iii Contents iv List of Figures vi List of Tables viii Chapter 1 Introduction 1 Chapter 2 Related Work 4 2.1 Single Image Super-resolution 4 2.1.1 Analytical Interpolation 4 2.1.2 Example-based Super-resolution 5 2.1.3 Learning-based Super-resolution. 6 2.2 Light-field Super-resolution 8 Chapter 3 The Focused Plenoptic Camera 10 3.1 Structure of the Focused Plenoptic Camera 11 3.2 Sampling Pattern of the Focused Plenoptic Camera 12 3.3 Rendering of the Focused Plenoptic Camera 16 3.3.1 Basic Rendering. 17 3.3.2 Rendering with Blending. 20 3.3.3 Multi-view Rendering. 22 3.3.4 Refocusing. 22 Chapter 4 Method 24 4.1 Single Image Super-resolution from Local Self-examples 25 4.2 Examples for Patches of Micro-images 27 4.2.1 Image Acquisition Process of the Focused Plenoptic Camera. 27 4.2.2 More Accurate Examples from Nearby Micro-images. 28 4.3 Estimating the Shift between Two Nearby Micro-images 30 4.3.1 Feature Detection. 31 4.3.2 Feature Matching and Voting. 31 4.4 Selection of the Four Example Candidates for Each Patch 32 Chapter 5 Results and Discussions 35 5.1 Input Data and Environment 35 5.2 Baselines 36 5.3 Visual Comparisons 36 5.4 Time Comparison 40 5.5 Discussions 40 Chapter 6 Conclusion 42 Bibliography 443116231 bytesapplication/pdf論文使用權限：同意有償授權(權利金給回饋本人)光場相機全場相機超解析度微影像範例thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/261484/1/ntu-103-R01922010-1.pdf