Neural image sharpening: a framework for volumetric wavefront sensing and imaging

This paper introduces a novel framework, to our knowledge, referred to here as neural image sharpening (NIS). NIS leverages implicit neural representations to sense and correct for anisoplanatic phase errors. In this paper, these phase errors manifest from the distributed-volume aberrations caused by atmospheric turbulence in digital-holographic (DH) measurements. With that said, NIS is not limited to this specific test case and has numerous potential benefits in applications where researchers gain access to the complex-valued optical field (e.g., microscopy, metrology, lidar, remote sensing, etc.). Nonetheless, using single-shot DH data, we employ NIS via unsupervised learning to perform volumetric wavefront sensing and imaging through turbulence (i.e., we sense and correct for the phase errors distributed along the turbulent propagation path). In so doing, we overcome the limitations of traditional pixel-based image sharpening (IS) frameworks. Using both wave-optics simulations and scaled-laboratory experiments, we demonstrate that NIS outperforms a traditional IS framework, achieving higher peak Strehl ratios over weak-, moderate-, and deep-turbulence conditions. Our findings establish NIS as a promising approach for volumetric wavefront sensing and imaging across numerous applications.