Optimal physical preprocessing for example-based super-resolution.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning Applications

Background:

• Fourier ptychographic microscopy (FPM) enables high space-bandwidth product imaging by computationally combining multiple low-resolution images acquired with varying illumination patterns.
• A key limitation of FPM is its poor temporal resolution due to the need for extensive image acquisition.
• Existing methods often struggle to balance imaging speed with resolution and field-of-view.

Purpose of the Study:

• To improve the temporal resolution of Fourier ptychographic microscopy for single-shot imaging.
• To achieve this without compromising the high space-bandwidth product characteristic of FPM.
• To integrate hardware modifications with deep learning for enhanced image reconstruction.

Main Methods:

• Employed example-based super-resolution techniques to learn the mapping from low-resolution to high-resolution images.
• Introduced physical preprocessing by modifying imaging hardware to capture more informative low-resolution data.
• Utilized deep learning to jointly optimize a single illumination pattern and post-processing reconstruction algorithm for specific sample types.

Main Results:

• Demonstrated that physical preprocessing significantly improves image reconstruction quality when combined with deep learning in FPM.
• Showcased that joint optimization of illumination and reconstruction yields superior results compared to optimizing only the reconstruction algorithm.
• Achieved improved image reconstruction in FPM through a synergistic approach of hardware modification and AI-driven algorithms.

Conclusions:

• Physical preprocessing is crucial for enhancing example-based super-resolution in Fourier ptychographic microscopy.
• Deep learning-based joint optimization of illumination and reconstruction offers a powerful strategy for single-shot, high-resolution imaging.
• This work establishes a new paradigm for accelerating FPM while maintaining its high imaging performance.