A Non-Local Low-Rank Approach to Enforce Integrability

Area of Science:

• Image processing and computer vision
• Applied mathematics and optimization
• Data science and machine learning

Background:

• Data corruption from outliers and dense noise poses significant challenges in various scientific and engineering fields.
• Existing methods for data restoration often struggle to simultaneously address both outlier corruption and dense noise.
• Advances in non-local methods offer new possibilities for robust data processing and signal recovery.

Purpose of the Study:

• To develop a new computational framework for enforcing data integrability.
• To design a regularization method that effectively handles sparse outliers and dense noise.
• To introduce an efficient optimization algorithm for solving the proposed formulation.

Main Methods:

• Formulation incorporating a sparse gradient data-fitting term to manage outliers.
• Integration of a gradient-domain non-local low-rank prior for regularization.
• Development of an efficient solver based on alternate minimization for the optimization problem.

Main Results:

• The proposed low-rank prior ensures patch similarity, aiding recovery from severe outlier corruption.
• The non-local low-rank prior demonstrates efficiency in reducing dense noise, similar to prior image restoration work.
• Experimental results show significant improvements over existing optimization methods in handling mixed noise and outliers.

Conclusions:

• The novel approach effectively enforces data integrability by combining sparse outlier handling with non-local low-rank regularization.
• The method provides a robust solution for data corrupted by both outliers and dense noise.
• The efficient alternate minimization solver enables practical application of the proposed technique for data restoration.