An Automatic Pixel-Wise Multi-Penalty Approach to Image Restoration.

Area of Science:

• Computer Vision
• Image Processing
• Applied Mathematics

Background:

• Image acquisition processes often introduce degradation like blur and noise.
• Effective image restoration is vital across various scientific and technical fields.
• Previous work utilized multi-penalty approaches based on the Uniform Penalty principle.

Purpose of the Study:

• To develop a new image restoration model and an iterative algorithm.
• To address the challenge of removing blur and noise from degraded images.
• To incorporate pixel-wise regularization and a parameter selection rule.

Main Methods:

• Developed a novel image restoration model incorporating pixel-wise regularization terms.
• Established a parameter selection rule for a sequence of constrained optimization problems.
• Adapted and proved the convergence of a modified Newton Projection method for multi-penalty scenarios.

Main Results:

• The proposed iterative algorithm effectively removes noise and blur from images.
• The method demonstrates a strong capability in preserving image edges during restoration.
• Numerical experiments validated the efficacy of the developed image restoration technique.

Conclusions:

• The new model and algorithm offer an effective solution for image restoration challenges.
• The approach successfully balances noise/blur removal with edge preservation.
• This work advances image restoration techniques with practical applications in science.