Simulated annealing, acceleration techniques, and image restoration

Area of Science:

• Computational Imaging
• Image Restoration
• Inverse Problems

Background:

• Linear image restoration is typically an ill-conditioned, underdetermined inverse problem.
• Stabilization often requires introducing smoothness constraints, which can impact edge preservation.
• Optimization of non-convex functionals is computationally challenging.

Purpose of the Study:

• To stabilize linear image restoration problems using a first-order smoothness constraint.
• To minimize a non-convex functional via stochastic relaxation with annealing.
• To investigate acceleration techniques for Metropolis-type annealing algorithms.

Main Methods:

• Employed stochastic relaxation with Metropolis dynamics for optimization.
• Introduced a first-order smoothness constraint to preserve edges.
• Investigated state space restriction and concave transform of the cost functional for acceleration.

Main Results:

• Developed and tested algorithms for space-variant restoration of synthetic aperture imaging data.
• Demonstrated significant benefits in convergence speed compared to standard Metropolis annealing.
• Achieved effective edge preservation and stabilization of the inverse problem.

Conclusions:

• The proposed acceleration techniques improve the practical performance of Metropolis annealing for image restoration.
• The method effectively addresses ill-conditioned inverse problems with significant convergence speed benefits.
• Successful application to synthetic aperture imaging data validates the algorithm's efficacy.