Learning Wavefront Coding for Extended Depth of Field Imaging.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning for Imaging

Background:

• Depth of field significantly impacts spatial information quality in imaging systems.
• Achieving an extended depth of field (EDoF) is a complex, ill-posed problem in optical imaging.
• Existing methods for EDoF often face limitations in performance and artifact reduction.

Purpose of the Study:

• To develop a computational imaging approach for achieving extended depth of field (EDoF).
• To jointly optimize optical design and image deblurring for enhanced imaging performance.
• To demonstrate superior EDoF imaging with minimal artifacts in challenging scenarios.

Main Methods:

• Employing wavefront coding with a diffractive optical element (DOE).
• Utilizing a convolutional neural network for image deblurring.
• Implementing end-to-end differentiable modeling for joint optimization of DOE and deblurring via gradient descent.
• Deriving an analytical expression for the DOE search space to ensure network convergence.

Main Results:

• Achieved superior EDoF imaging performance compared to current state-of-the-art methods.
• Demonstrated minimal artifacts in acquired images across various scenarios.
• Successfully imaged deep 3D scenes and performed broadband imaging with extended depth of field.

Conclusions:

• The proposed computational imaging approach effectively enhances depth of field.
• Joint optimization of optical elements and deep learning models offers a powerful strategy for EDoF imaging.
• This method shows significant potential for applications requiring high-quality imaging over extended depth ranges.