Image super-resolution using dual regression network and kernel estimation
Image super-resolution (SR) technology has always been an important research direction in the field of image processing. In recent years, deep learning technology has been applied to deal with SR problems, and many effective network models have been designed. However, deep learning algorithms are...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2021
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/150214 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Image super-resolution (SR) technology has always been an important research direction in the field of image processing. In recent years, deep learning technology has been applied to deal with SR problems, and many effective network models have been designed.
However, deep learning algorithms are a data-driven method, and SR networks require pairs of low-resolution (LR) and high-resolution (HR) images as training data. At present, most of the training data sets adopted by SR networks are generated by a simple bicubic method, which does not meet the requirements of processing real-world photos. Therefore, this project will solve this problem by constructing new training and testing datasets by estimating the degradation convolution kernel from real-world images.
In this project, we will build a new training data set based on the Kernel GAN method and apply it to the latest network model, and then perform a SR test on the benchmark dataset and evaluate its performance. |
|---|