Candidate:
Zhuoran
Li
Title:
Towards
Accurate
Quality
Control
of
Visual
Coding:
An
Eigen
Analysis
Approach
Date:
July
21,
2022
Time:
12:00
Place:
online
Supervisor(s):
Wang,
Zhou
Abstract:
While pushing the rate-quality (RD) limit to the next level will always be the paramount goal of video/image compression models, precise control in terms of both rate and quality is the necessary cornerstone for real applications. Recent years have witnessed great advancement of accurate rate control method for video encoders, whereas quality control that based on the accurate video quality and encoding parameter modelling is less visited. In this paper, we construct and analyze the generalized function space of quality-encoding parameter. We then propose an eigen analysis approach for the modelling of image/video quality against the encoding parameter, which is named as generalized quality parameter (GQP) model. The theoretical function space is defined and proved to be a convex set in a Hilbert space, which inspires a computational model of GQP function and a method of sparse measurements parameter estimation. Two large-scale databases, one video and the other image, are used to demonstrate the feasibility of our idea through experiments. With the computational model and the sparse measurements method, the GQP function of a specific video/image can be reconstructed accurately from only a few queries, which significantly outperforms the current widely used empirical estimation methods both in accuracy and efficiency.