A first-order splitting method for solving a large-scale composite convex optimization problem

TL;DR

This paper introduces new first-order splitting algorithms for large-scale composite convex optimization problems, effectively handling multiple convex functions, including nonsmooth and smooth components, with applications in CT image reconstruction.

Contribution

The paper develops several efficient, matrix-inversion-free, splitting algorithms for multi-block convex problems, extending existing methods to more complex composite functions.

Findings

Algorithms outperform existing methods in numerical tests.

Effective in solving CT image reconstruction problems.

Handle multiple convex functions with different properties.

Abstract

The forward-backward operator splitting algorithm is one of the most important methods for solving the optimization problem of the sum of two convex functions, where one is differentiable with a Lipschitz continuous gradient and the other is possibly nonsmooth but proximable. It is convenient to solve some optimization problems in the form of dual or primal-dual problems. Both methods are mature in theory. In this paper, we construct several efficient first-order splitting algorithms for solving a multi-block composite convex optimization problem. The objective function includes a smooth function with a Lipschitz continuous gradient, a proximable convex function that may be nonsmooth, and a finite sum of a composition of a proximable function and a bounded linear operator. To solve such an optimization problem, we transform it into the sum of three convex functions by defining an appropriate inner product space. On the basis of the dual forward-backward splitting algorithm and the primal-dual forward-backward splitting algorithm, we develop several iterative algorithms that involve only computing the gradient of the differentiable function and proximity operators of related convex functions. These iterative algorithms are matrix-inversion-free and completely splitting algorithms. Finally, we employ the proposed iterative algorithms to solve a regularized general prior image constrained compressed sensing (PICCS) model that is derived from computed tomography (CT) image reconstruction under sparse sampling of projection measurements. Numerical results show that the proposed iterative algorithms outperform other algorithms.