Image Restoration by Iterative Denoising and Backward Projections

Area of Science:

• Computational Imaging
• Optimization Theory
• Applied Mathematics

Background:

• Inverse problems are prevalent in image deblurring and inpainting.
• Traditional methods require problem-specific algorithms.
• The Plug-and-Play (P&P) framework utilizes denoising algorithms but often needs extensive parameter tuning.

Purpose of the Study:

• To propose a novel method for solving inverse problems with reduced parameter tuning.
• To leverage off-the-shelf denoisers for general inverse problems.
• To develop an automatic parameter tuning mechanism.

Main Methods:

• Transformed a standard cost function into a novel optimization problem.
• Developed an efficient minimization scheme with Plug-and-Play properties.
• Integrated an automatic parameter tuning mechanism.

Main Results:

• The proposed method requires less parameter tuning compared to existing P&P approaches.
• Empirical results demonstrate competitiveness with task-specific methods.
• Successfully applied to image inpainting and deblurring tasks.

Conclusions:

• The new method offers an efficient and less parameter-intensive alternative for solving inverse problems.
• It effectively utilizes existing denoising algorithms.
• Provides a viable solution for image restoration applications.