Efficient algorithms for robust recovery of images from compressed data.

Area of Science:

• Signal Processing
• Optimization Theory

Background:

• Compressed Sensing (CS) enables sub-Nyquist sampling and recovery of compressible data.
• Robust Compressed Sensing (robust CS) addresses data corruption from impulsive noise by integrating robust statistics.
• Existing robust CS algorithms are computationally inefficient, relying on iterative solutions of standard CS problems.

Purpose of the Study:

• To develop more computationally efficient algorithms for robust CS.
• To extend the robust CS framework to handle more complex scenarios.
• To demonstrate the improved performance and broader applicability of the new methods.

Main Methods:

• Leveraging advances in large-scale convex optimization for nonsmooth regularization.
• Developing iterative algorithms based on efficient optimization techniques.
• Extending the robust CS formulation with affine constraints, L1-norm loss, mix-norm regularization, and multitasking.

Main Results:

• The proposed algorithms offer significant computational advantages over the original robust CS method.
• New algorithms effectively solve sophisticated robust CS extensions not addressable by prior methods.
• Demonstrated utility of extended robust CS formulations in various imaging tasks.

Conclusions:

• The developed algorithms provide a computationally efficient and effective solution for robust CS.
• The extended robust CS framework enhances data recovery in complex and diverse settings.
• The new methods offer practical improvements for imaging and other applications.