Robust image deblurring with an inaccurate blur kernel.

Area of Science:

• Computer Vision
• Image Processing
• Optimization

Background:

• Nonblind image deblurring methods often assume accurate blur kernels.
• Real-world blur kernels can contain significant errors, leading to image distortions.
• Nonuniform motion blurring is a common source of severe blur kernel inaccuracies.

Purpose of the Study:

• To develop a robust image deblurring method that explicitly handles errors in the blur kernel.
• To improve image restoration quality when the blur kernel is not perfectly known.
• To enhance the resilience of deblurring algorithms to both noise and kernel inaccuracies.

Main Methods:

• A novel convex minimization model is proposed to incorporate blur kernel error.
• The accelerated proximal gradient method is employed for efficient problem solving.
• A new boundary extension scheme is integrated to refine deblurring results.

Main Results:

• The proposed method demonstrates efficiency and robustness against image noise.
• The algorithm effectively mitigates distortions caused by erroneous blur kernels.
• Experiments on synthetic and real images validate the approach's performance.

Conclusions:

• The developed convex minimization model offers a robust solution for nonblind image deblurring with inaccurate kernels.
• The integration of a boundary extension scheme further improves restoration quality.
• The algorithm shows significant advantages over existing methods when dealing with real-world image degradations.