Abstract
Quantitative phase imaging (QPI), as a label-free microscopic imaging technique, provides quantitative optical information of cells and their subcellular structures, holding significant value in biomedical research. However, existing methods still face a notable trade-off between resolution and data efficiency. Differential phase contrast (DPC) imaging is constrained by the diffraction limit under partially coherent illumination, with its frequency support typically not exceeding twice the objective numerical aperture ( NAobj ). Although Fourier ptychographic microscopy (FPM) can achieve higher resolution, it relies on a large number of multi-angle images, resulting in high costs for data acquisition and reconstruction. To address these challenges, we propose an unsupervised super-resolution quantitative phase imaging (SRQPI) method, which reconstructs the phase using two orthogonal annular bright-field measurements matched to the NAobj and one high-angle dark-field measurement with illumination extended to 2NAobj. By embedding the physical forward model of partially coherent imaging into the network and leveraging the implicit regularization constraints inherent in the network architecture, the approach achieves stable phase reconstruction without any external training data. Benefiting from the synergistic fusion of DPC and dark-field information, the method effectively expands the spectral coverage, enabling equivalent 3NAobj imaging capability. In USAF resolution target simulation, the proposed SRQPI method significantly improves spatial resolution compared with traditional DPC using only three images. Spectral analysis further demonstrates that the method breaks through the conventional 2NAobj frequency limit of partially coherent imaging. These results suggest that the proposed DIP-based SRQPI framework provides a data-efficient route toward high-resolution QPI.