Pixel-super-resolved lensless on-chip microscopy via precision-enhanced dynamic registration and complex-constrained phase retrieval

Lensless on-chip microscopy (LOCM) has emerged as a research focus in computational imaging due to its compact structure and high-throughput imaging capabilities. However, precise calibration of misalignments between axial holograms is challenging, and these errors accumulate and magnify during the phase retrieval process, limiting the imaging resolution and detail fidelity. We propose a pixel-super-resolved LOCM via precision-enhanced dynamic registration and complex-constrained phase retrieval. The approach incorporates a dynamic registration mechanism in the iterative reconstruction process, effectively suppressing global misalignment errors. The introduction of physical constraints improves the convergence stability and further enhances the fidelity. Experimental results across different sample types demonstrate a 1.78-fold improvement in spatial resolution beyond the Nyquist-Shannon sampling limit across a full field of view of 28.6 mm², with the minimum resolvable size reaching 775 nm. This method achieves superior performance in resolving intricate structural details, offering an effective solution for high-fidelity lensless imaging.