Nested Conjugate Gradient Algorithm With Nested Preconditioning for Non-Linear Image Restoration.

Area of Science:

• Computational mathematics
• Image processing
• Optimization algorithms

Background:

• Image restoration is crucial for enhancing degraded image quality.
• Existing optimization methods face challenges with complex regularization functionals.
• Efficient algorithms are needed for both convex and non-convex image restoration tasks.

Purpose of the Study:

• To develop a novel and efficient optimization algorithm for image restoration.
• To address limitations of current methods in handling quadratic data fitting and smooth non-quadratic regularization.
• To provide a robust solution for both convex and non-convex regularization scenarios.

Main Methods:

• Development of the nested non-linear conjugate gradient (NNCG) algorithm.
• NNCG algorithm combines outer non-linear CG with an inner linear CG preconditioning.
• Inner CG iteration is accelerated using an FFT-based non-iterative preconditioner.
• Theoretical convergence proof to a stationary point for convex and non-convex functionals.

Main Results:

• The NNCG algorithm demonstrates superior performance compared to established methods.
• Experimental results show outperformance against majorization-minimization for convex regularization.
• NNCG surpasses non-convex inertial-proximal methods for non-convex regularization.
• The method achieves convergence for both convex and non-convex regularization functionals.

Conclusions:

• The proposed NNCG algorithm is an effective and efficient method for image restoration.
• NNCG offers significant improvements over existing optimization techniques.
• The algorithm provides a unified approach for both convex and non-convex regularization problems in image restoration.