A new approach for atmospheric turbulence removal using low-rank matrix factorization

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Atmospheric turbulence significantly degrades image quality, introducing geometric distortions and spatiotemporal blur.
• Existing methods often struggle to address both distortion types simultaneously and effectively.

Purpose of the Study:

• To present a novel iterative method for removing atmospheric turbulence from image sequences.
• To restore high-quality, clear images from turbulent inputs by modeling distortion and blur.

Main Methods:

• Modeling turbulence using geometric pixel transformation and spatiotemporal varying blur.
• Employing low-rank matrix factorization within an iterative framework.
• Utilizing random sample consensus for image subset selection and local transformation matrix estimation.

Main Results:

• Successfully mitigated substantial geometrical distortions in turbulent images.
• Improved spatiotemporal varying blur, enhancing image clarity.
• Effectively restored fine details lost due to atmospheric turbulence.

Conclusions:

• The proposed low-rank matrix factorization method offers a robust solution for atmospheric turbulence removal.
• The technique demonstrates significant improvements in image quality restoration for both real and simulated data.
• This approach effectively handles complex distortions and blur, leading to high-fidelity image reconstruction.