A scaled, inexact and adaptive Fast Iterative Soft-Thresholding Algorithm for convex image restoration

TL;DR

This paper introduces an adaptive, inexact variant of the Fast Iterative Soft-Thresholding Algorithm (FISTA) tailored for convex image restoration, featuring automatic step-size adjustment and quadratic convergence, validated on TV image deblurring with Poisson noise.

Contribution

It proposes a novel adaptive, inexact FISTA algorithm for convex image restoration with automatic step-size tuning and proven quadratic convergence.

Findings

Improved computational efficiency in image deblurring tasks.

Enhanced precision in restoring images with Poisson noise.

Validated quadratic convergence depending on backtracking parameters.

Abstract

In this note, we consider a special instance of the scaled, inexact and adaptive generalised Fast Iterative Soft-Thresholding Algorithm (SAGE-FISTA) recently proposed in (Rebegoldi, Calatroni, '21) for the efficient solution of strongly convex composite optimisation problems. In particular, we address here the sole (non-strongly) convex optimisation scenario, which is frequently encountered in many imaging applications. The proposed inexact S-FISTA algorithm shows analogies to the variable metric and inexact version of FISTA studied in (Bonettini, Rebegoldi, Ruggiero, '19), the main difference being the use of an adaptive (non-monotone) backtracking strategy allowing for the automatic adjustment of the algorithmic step-size along the iterations (see (Scheinberg, Goldfarb, Bai, '14, Calatroni, Chambolle, '19)). A quadratic convergence result in function values depending on the backtracking parameters and the upper and lower bounds on the spectrum of the variable metric operators is given. Experimental results on TV image deblurring problems with Poisson noise are then reported for numerical validation, showing improved computational efficiency and precision.