Towards High Fidelity Image and Video Restoration

PhD Thesis Proposal Defence


Title: "Towards High Fidelity Image and Video Restoration"

by

Mr. Hao OUYANG


Abstract:

As multimedia content becomes increasingly ubiquitous in our daily lives, 
there is a growing demand for restoring sequences from multiple domains, 
as well as high-resolution images and videos from the real world. The goal 
of image and video restoration is to repair corrupted regions of visual 
data by replacing them with content that is both visually realistic and 
semantically consistent. This type of algorithm can be beneficial for 
general users in a variety of practical applications, such as removing 
unwanted objects, repairing facial defects, and editing data. Despite 
significant progress in restoration techniques, there are still issues 
with abrupt color artifacts and severe flickering in the resulting images 
or videos.

The objective of the thesis is to enhance the performance of image and 
video restoration using internal learning techniques and flow-based 
models. To improve image restoration, we utilize a unique 
external-internal inpainting strategy with a monochromic bottleneck. 
Additionally, we suggest an alternative method to minimize corruption 
caused by editing through learning a restorable operator using 
quasi-invertible deep models. However, the image restoration algorithm 
doesn't consider temporal information and 3D structure, leading to severe 
flickering when applied to videos. We propose that the information 
propagation process can be implicitly addressed by cross-frame correlation 
and enforcing gradient constraints for video restoration. Furthermore, we 
introduce 3D representations for high-fidelity inpainting of novel views. 
These proposed approaches have demonstrated satisfactory performance on 
various image and video restoration tasks.


Date:			Friday, 5 May 2023

Time:                  	4:00pm - 6:00pm

Venue:			Room 5501
  			lifts 25/26

Committee Members:	Dr. Qifeng Chen (Supervisor)
 			Dr. Dan Xu (Chairperson)
 			Prof. Chiew-Lan Tai
 			Prof. Chi-Keung Tang


**** ALL are Welcome ****