A primal-dual active-set method for non-negativity constrained total variation deblurring problems

Area of Science:

• Image Processing
• Numerical Analysis
• Optimization

Background:

• Image deblurring is crucial for image restoration.
• Total variation (TV) models enhance image quality but pose computational challenges.
• Non-negativity constraints improve solution fidelity but complicate algorithms.

Purpose of the Study:

• Develop a fast and robust numerical algorithm for non-negatively constrained total variation-based image deblurring.
• Address the computational difficulties introduced by the non-negativity constraint.
• Improve the efficiency and accuracy of solving constrained deblurring problems.

Main Methods:

• Formulated the constrained deblurring problem as a primal-dual program, adapting existing unconstrained formulations.
• Employed a semi-smooth Newton's method to solve the primal-dual program.
• Leveraged the connection between semi-smooth Newton and primal-dual active set methods for computational simplification.

Main Results:

• The proposed algorithm demonstrates a quadratic rate of local convergence and numerical evidence of global convergence.
• Achieved high accuracy in solving the optimality system.
• Showcased robustness across a wide range of parameters with minimal parameter adjustment.
• Numerical comparisons confirm speed and accuracy advantages over existing methods.

Conclusions:

• The developed semi-smooth Newton's method offers an efficient and accurate solution for non-negatively constrained image deblurring.
• The algorithm's robustness and convergence properties make it a valuable tool for image restoration tasks.
• This work advances the state-of-the-art in solving complex image deblurring problems.