Abstract:
CNNs have been used to study the single image super-resolution problem. Most CNN-based methods learn a model to map a downsampled low-resolution (LR) image to a high-resolution (HR) image. In general, when the down-sampling process is unknown and the LR input is degraded by noise and blurring, it is difficult to acquire the LR and HR image pairs for traditional supervised learning.
Inspired by recent unsupervised imagestyle translation applications using unpaired data, we propose a multiple Cycle-in-Cycle network structure with multiple generative adversarial networks (GAN) as basis components to handle the more general case.
The first network cycle maps the noisy and blurry LR input to a noise-free LR space, then a new cycle with a well-trained x2 network model super-resolves the intermediate output of the first cycle.
Up-sampling factors (x2, x4, x8) determine the total cycles. To achieve HR output, all modules are trained end-to-end. Quantitative and qualitative results show that our proposed method performs similarly to state-of-the-art supervised models.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
