Fast, robust total variation-based reconstruction of noisy, blurred images

Area of Science:

• Image processing and computational imaging.
• Applied mathematics and numerical analysis.

Background:

• Image degradation is a common problem in digital imaging, caused by noise and blurring.
• Traditional image restoration methods often impose a priori smoothness constraints, which may not be suitable for all image types.

Purpose of the Study:

• To develop and evaluate an image reconstruction technique for noisy and blurred data.
• To apply Tikhonov regularization with a modified total variation (TV) functional for image recovery.
• To present an efficient algorithm for the discretized problem.

Main Methods:

• Utilized Tikhonov regularization with a modified total variation regularization functional.
• Developed an efficient algorithm combining fixed point iteration for nonlinearity and preconditioned conjugate gradient iteration for linear systems.
• Applied the method to a satellite image reconstruction problem.

Main Results:

• Successfully reconstructed images from noisy, blurred data.
• Demonstrated that the method does not require a priori smoothness conditions on the solution image.
• Presented convergence results and a direct comparison with a fast linear solver, showing effectiveness.

Conclusions:

• The proposed Tikhonov regularization with modified TV functional is effective for image reconstruction from degraded data.
• The efficient algorithm presented is suitable for practical image processing applications, including satellite imagery.
• The method offers an advantage over techniques requiring a priori smoothness assumptions.