Date of Award

June 2018

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Mathematics

Advisor(s)

Yuesheng Xu

Second Advisor

Lixin Shen

Keywords

Convex Optimization, Fixed-point Algorithm, Image Processing, Proximity Operator

Subject Categories

Physical Sciences and Mathematics

Abstract

A variety of optimization problems especially in the field of image processing are not differentiable in nature. The non-differentiability of the objective functions together with the large dimension of the underlying images makes minimizing the objective function theoretically challenging and numerically difficult. The fixed-point proximity framework that we will systematically study in this dissertation provides a direct and unified methodology for finding solutions to those optimization problems. The framework approaches the models arising from applications straightforwardly by using various fixed point techniques as well as convex analysis tools such as the subdifferential and proximity operator.

With the notion of proximity operator, we can convert those optimization problems into finding fixed points of nonlinear operators. Under the fixed-point proximity framework, these fixed point problems are often solved through iterative schemes in which each iteration can be computed in an explicit form. We further explore this fixed point formulation, and develop implicit iterative schemes for finding fixed points of nonlinear operators associated with the underlying problems, with the goal of relaxing restrictions in the development of solving the fixed point equations. Theoretical analysis is provided for the convergence of implicit algorithms proposed under the framework. The numerical experiments on image reconstruction models demonstrate that the proposed implicit fixed-point proximity algorithms work well in comparison with existing explicit fixed-point proximity algorithms in terms of the consumed computational time and accuracy of the solutions.

Access

Open Access

Share

COinS