A new difference of anisotropic and isotropic total variation regularization method for image restoration

Area of Science:

• Image Processing
• Computer Vision
• Applied Mathematics

Background:

• Total variation (TV) regularization is widely used in image processing for its ability to preserve edges.
• Existing TV methods often struggle with non-smoothness and convexity, limiting their performance in complex restoration tasks.
• Developing advanced regularization techniques is crucial for improving image quality in various applications.

Purpose of the Study:

• To propose a new nonconvex total variation regularization method for image restoration.
• To address the challenges of non-convexity and non-smoothness in TV-based image restoration.
• To enhance the performance of image restoration algorithms using a novel mathematical framework.

Main Methods:

• A novel nonconvex total variation regularization model is developed, incorporating the generalized Fischer-Burmeister function.
• Specific difference of convex algorithms (DCA) are designed to handle the non-convex and non-smooth nature of the proposed model.
• The subproblem within the DCA is efficiently minimized using the alternating direction method of multipliers (ADMM).

Main Results:

• The proposed DCA-ADMM algorithms exhibit low computational complexity per iteration.
• Experimental results in image denoising show improved performance compared to state-of-the-art methods.
• Applications in magnetic resonance imaging (MRI) also demonstrate the effectiveness and preference for the proposed restoration models.

Conclusions:

• The proposed nonconvex TV regularization method offers a significant advancement in image restoration.
• The developed DCA-ADMM approach provides an efficient and effective solution for complex image restoration problems.
• This work contributes a valuable tool for enhancing image quality in fields like medical imaging and computer vision.