Linearly involved Generalized Moreau Enhanced Model with Non-quadratic Smooth Convex Data Fidelity Functions
Wataru Yata, Keita Kume, Isao Yamada
Published: 2025/9/3
Abstract
In this paper, we introduce an overall convex model incorporating a nonconvex regularizer. The proposed model is designed by extending the least squares term in the constrained LiGME model [Yata Yamagishi Yamada 2022] to fairly general smooth convex functions for flexible utilization of non-quadratic data fidelity functions. Under an overall convexity condition for the proposed model, we present sufficient conditions for the existence of a minimizer of the proposed model and an inner-loop free algorithm with guaranteed convergence to a global minimizer of the proposed model. To demonstrate the effectiveness of the proposed model and algorithm, we conduct numerical experiments in scenarios of Poisson denoising problem and simultaneous declipping and denoising problem.