Computational microscopy is vital in biomedicine and materials science. Traditional methods struggle with optical aberrations, noise interference, and differences between physical models and real-world imaging, reducing the resolution and accuracy. They rely on pixel-level optimization, which fails to maintain high-quality imaging in complex environments. Therefore, developing a precise and stable computational imaging approach has become a research focus.

In a study published in Advanced Science, Prof. PAN An from Xi’an Institute of Optics and Precision Mechanics (XIOPM) of the Chinese Academy of Sciences and Prof. CAO Liangcai’s team from Tsinghua University developed a new method, the feature-domain phase retrieval (FD-PR), which significantly improves imaging resolution and interference resistance, representing a breakthrough in computational microscopy.

The proposed FD-PR algorithm differs from traditional pixel-based methods. This algorithm gets stable features such as image edges and textures to create a feature domain loss function for wavefront reconstruction. It effectively handles noise issues, with enhanced robustness and adaptability. Besides, the FD-PR framework includes a flexible constraint block, supporting different adjustable constraints, further expanding its applicability.

Researchers found that the FD-PR performs exceptionally well in multiple applications. In full-field Fourier ptychography, the method effectively eliminated the vignetting effect, enabling the large field of view, and high-resolution imaging. In noise-free coded ptychography experiments, it surpassed the traditional algorithm in reducing noise, enhancing contrast, and improving reconstruction quality. For inline holography, it combined physical limits with image denoising techniques, achieving high-quality phase recovery under single exposure. In blind large aberrations recovery, it recovers aberrations up to 6π.

“The FD-PR provides a universal and efficient phase recovery framework for computational microscopy, and it offers strong adaptability and robustness to support the development of next-generation imaging technologies,” said Prof. PAN An