A variant of two-step modulus-based matrix splitting iteration method for Retinex problem

作     者：Chen, Fang Zhu, Yu 

作者机构：Beijing Informat Sci & Technol Univ Sch Appl Sci Beijing 100192 Peoples R China 

出 版 物：《COMPUTATIONAL & APPLIED MATHEMATICS》 (Comput. Appl. Math.)

年 卷 期：2022年第41卷第6期

页      面：1-13页

核心收录：

基　　金：R&D Program of Beijing Municipal Education Commission [KM201911232010] 

主　　题：Linear complementarity problem Two-step iteration method Modulus-based matrix splitting Retinex problem 

摘      要：Based on a variational optimization model, and by imposing physical constraints on the reflection value, and deriving deformation of the Retinex problem, we find that the Retinex problem is equivalent to a linear complementarity problem and its solution can be computed by solving an equivalent fixed-point equation. In light of the theoretical analysis of the special structure of the system matrix of the linear complementarity problem, we propose a variant of the two-step modulus-based matrix splitting iteration method, and then prove its unconditional convergence. We further give practically quasi-optimal values of the involved iteration parameters in this method. The numerical results show that the variant of the two-step modulus-based matrix splitting iteration method is effective in terms of iteration steps, computing time, and natural image quality evaluator.