| 摘要 | 第3-4页 |
| Abstract | 第4页 |
| 1 Introduction | 第7-26页 |
| 1.1 Research Background | 第7-9页 |
| 1.2 Research Objective and Significance | 第9-11页 |
| 1.2.1 Research Objective | 第9-10页 |
| 1.2.2 Research Significance | 第10-11页 |
| 1.3 Domestic and Overseas Progress | 第11-22页 |
| 1.3.1 Research progress of super – resolution technologies | 第11-18页 |
| 1.3.2 State and perspectives of machine learning | 第18-20页 |
| 1.3.3 Overseas progress | 第20-22页 |
| 1.4 Main Content and Research Methods | 第22-26页 |
| 1.4.1 Main Content | 第22-23页 |
| 1.4.2 Research Method | 第23-26页 |
| 2 Theoretical and model Analysis | 第26-39页 |
| 2.1 General statement of the super-resolution problem | 第26-29页 |
| 2.1.1 Common Image distorted model | 第26-27页 |
| 2.1.2 The model of distortion process | 第27页 |
| 2.1.3 The distortion functions | 第27-29页 |
| 2.2 Main methods for solving super-resolution problem | 第29-34页 |
| 2.2.1 The classical methods | 第29-30页 |
| 2.2.2 Methods that use a-priori information | 第30-31页 |
| 2.2.3 Methods that explicitly modeling a non-linearity | 第31-32页 |
| 2.2.4 Neural Network Method | 第32-34页 |
| 2.3 The applicability of neural networks for super-resolution problem | 第34-39页 |
| 2.3.1 Overview of Deep Convolutional Neural networks | 第34-35页 |
| 2.3.2 Realization of generative adversarial networks | 第35-39页 |
| 3 Image Reconstruction Technology | 第39-45页 |
| 3.1 Instruments for super-resolution image restoration | 第39-41页 |
| 3.1.1 Image quality metrics | 第39页 |
| 3.1.2 Calculation of harmonic lenses | 第39-40页 |
| 3.1.3 Image registration using a single harmonic lens | 第40-41页 |
| 3.2 Instruments for super-resolution image restoration | 第41-45页 |
| 3.2.1 Image quality metrics | 第41-43页 |
| 3.2.2 Evaluation of the point spread function (PSF) | 第43-44页 |
| 3.2.3 Elimination of chromatic distortion based on deconvolution | 第44-45页 |
| 4 Experiments | 第45-66页 |
| 4.1 The first approach: Combination of GAN and DCNN | 第45-54页 |
| 4.1.1 Application of GAN | 第45-48页 |
| 4.1.2 The first experiment: The use of error function | 第48-52页 |
| 4.1.3 The second experiment: The use of hidden representations | 第52-54页 |
| 4.2 The error function generated by a deep neural network | 第54-56页 |
| 4.3 Third approach: Elimination of distortion based on DCNN | 第56-60页 |
| 4.3.1 The interchannel communication | 第56-57页 |
| 4.3.2 DCNN Building | 第57-60页 |
| 4.4 Perfomance Evaluation | 第60-66页 |
| 4.4.1 The parameters of conducted researches | 第60-61页 |
| 4.4.2 Comparative study of reconstruction quality | 第61-64页 |
| 4.4.3 Comparison of the DCNN and GPU | 第64-66页 |
| Conclusions and Future Work | 第66-69页 |
| Conclusion | 第66-67页 |
| Future Work | 第67-69页 |
| References | 第69-74页 |
| Research Projects and Publications in Master Study | 第74-75页 |
| Acknowledgement | 第75-78页 |
