Diffusion based burst super resolution

Date

2024-1-25

Author

Cengiz, Ali Baran

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

435
views

883
downloads

Cite This

Current diffusion modeling approaches generate high-quality images by condition- ing on a single image. However, super-resolution is an inverse problem, with multi- ple possible correct reconstructions, lacking a unique solution. In order to converge generated images closer to the actual high-resolution image, multiple input images captured in a quick burst are used in the burst super-resolution literature. This thesis proposes several ways to adapt diffusion models to leverage multiple inputs, thereby potentially improving their reconstruction quality. In addition, inherent challenges in both the super-resolution problem and proposed adaptations are also examined in detail, and possible solutions are discussed.

Subject Keywords

Super resolution, Image processing, Burst imaging, Generative models

URI

https://hdl.handle.net/11511/108743

Collections

Graduate School of Natural and Applied Sciences, Thesis