Blind Image Inpainting with Mixture Noise Using ℓ ₀ and Total Regularization

Area of Science:

• Medical Imaging
• Computer Vision
• Optimization

Background:

• Blind image inpainting is challenging, especially with large datasets like medical images.
• Existing methods struggle with the complexity of reconstructing missing image data.

Purpose of the Study:

• To develop an effective method for blind image inpainting using a nonconvex sparse optimization model.
• To address the challenges posed by large-scale medical image datasets.

Main Methods:

• A proximal based alternating direction method of multipliers (PADMM) was designed.
• Incorporated ℓ₀ sparse regularization for binary masks.
• Utilized total variation and ℓ₂ regularization for image reconstruction.

Main Results:

• The proposed PADMM method demonstrated superior performance compared to traditional approaches.
• Successfully handled blind image inpainting tasks, including those in medical imaging.

Conclusions:

• The developed nonconvex sparse optimization model with PADMM is effective for blind image inpainting.
• This method offers a significant improvement for reconstructing missing data in medical images.