Wide-Field, Artifact-Suppressed Lensless Microscopy Via Feature-Agnostic Autofocus and Multi-Target Separable Phase Retrieval

Lensless on-chip microscopy (LOCM) is a promising technique for high-throughput, label-free imaging. However, its practical implementation remains constrained by sensitivity to variations in sample-sensor distance and cumulative noise during reconstruction. Existing autofocus and phase retrieval algorithms fail when dealing with non-planar samples and environmental disturbances. A wide-field, artifact-suppressed lensless imaging framework with natural support constraints is proposed, which integrates segment-dependent lateral registration, feature-agnostic autofocus, and multi-target separable phase retrieval. The proposed complex field correction scheme effectively addresses diffraction scale distortion and sensor trajectory deviations while mitigating the dependence of autofocus on sample-specific image features. The phase retrieval process is further enhanced via a multi-objective stochastic gradient descent algorithm, which enables effective noise separation without compromising resolution. Experimental validations with diverse samples demonstrate significant performance improvements, achieving pixel-super-resolved imaging with a full field of view (FOV) of 28.6 ${\mathrm{mm}}^2$, an imaging depth exceeding 80 $\mu \mathrm{m}$, and a half-pitch resolution of 775 nm, corresponding to a 2.15-fold improvement in spatial resolution beyond the Nyquist–Shannon sampling limit. Furthermore, the proposed method demonstrates strong compatibility with coherent diffraction imaging (CDI), highlighting its broader applicability. The proposed method improves the versatility of lensless microscopy, eliminates reliance on stringent calibration, and provides a robust solution for high-throughput computational imaging.