Abstract:The proximal gradient method is a generic technique introduced to tackle the non-smoothness in optimization problems, wherein the objective function is expressed as the sum of a differentiable convex part and a non-differentiable regularization term. Such problems with tensor format are of interest in many fields of applied mathematics such as image and video processing. Our goal in this paper is to address the solution of such problems with a more general form of the regularization term. An adapted iterative proximal gradient method is introduced for this purpose. Due to the slowness of the proposed algorithm, we use new tensor extrapolation methods to enhance its convergence. Numerical experiments on color image deblurring are conducted to illustrate the efficiency of our approach.
What problem does this paper attempt to address?
This paper mainly focuses on global approximate gradient methods for multidimensional optimization problems, especially for nonsmooth regularization problems in image processing. Problem (1.1) is a constrained nonsmooth optimization problem with tensor format, where the objective function is composed of differentiable convex functions and nonsmooth regularization terms. The paper proposes an adaptive iterative approximate gradient method to handle the nonsmoothness of the regularization term, and accelerates the convergence of the algorithm by introducing new tensor extrapolation techniques.
Specifically, the paper first introduces the background of the problem, including nonsmooth optimization and its applications in image processing. Then, the authors propose a global bi-approximate gradient method that uses a tensor form of the objective function, and discuss its special cases such as l1 regularization, l1-l2 regularization, and total variation regularization. Due to the slow convergence of basic gradient methods, the paper introduces two tensor extrapolation techniques to improve algorithm efficiency.
Next, the paper describes the algorithm steps in detail, including parameter selection, the use of Tseng's splitting algorithm for handling constrained problems, and how to project the solution onto the constraint set through Tseng's splitting algorithm. In addition, the paper also discusses the strategies for choosing parameters αk and βl to ensure the convergence of the sequence.
Finally, the paper demonstrates the effectiveness of the algorithm in the application of color image deblurring through numerical experiments, and compares it with several accelerated versions of algorithms, including GTPG algorithm using GT-TET (Global Tensor Topology Extrapolation Transformation) and HOSVD-MPE (Higher-order Singular Value Decomposition based on Minimum Extrapolation) acceleration.
In conclusion, this paper proposes a general algorithm for dealing with nonsmooth tensor optimization problems and improves the convergence speed of the algorithm by introducing extrapolation techniques, which is particularly suitable for problems in image processing.