Impulse Noise Image Restoration Using Nonconvex Variational Model and Difference of Convex Functions Algorithm.

Area of Science:

• Image processing
• Computer vision
• Applied mathematics

Background:

• Impulse noise significantly degrades image quality.
• Traditional L1 norm and total variation (TV) regularization methods create staircase artifacts and are not robust to impulse noise.
• Existing convex optimization methods struggle with effective impulse noise removal.

Purpose of the Study:

• To develop a novel variational model for high-quality impulse noise image restoration.
• To address the limitations of existing methods, specifically staircase artifacts and poor noise robustness.
• To introduce a new algorithm for solving the proposed nonconvex variational model.

Main Methods:

• Proposed a variational model integrating nonconvex data fitting and nonconvex total variation (TV) regularization.
• Utilized a nonconvex TV regularizer to eliminate staircase artifacts.
• Employed a nonconvex fidelity term for effective impulse noise detection, distinguishing between slightly and severely corrupted pixels.
• Developed a novel difference of convex functions algorithm to solve the variational model.
• Proved the convergence of the proposed algorithm to a stationary point.

Main Results:

• The nonconvex TV regularizer successfully eliminated staircase artifacts.
• The nonconvex fidelity term demonstrated effective impulse noise detection.
• The developed algorithm proved efficient and converged to a stationary point.
• Experimental results showed superior performance compared to state-of-the-art methods.

Conclusions:

• The proposed variational model with nonconvex regularization effectively restores images corrupted by impulse noise.
• The novel algorithm provides an efficient solution for the nonconvex optimization problem.
• The method achieves high-quality image restoration, outperforming existing techniques.