End-to-end super-resolution computational imaging with super-oscillatory light field.

Super-oscillatory (SO) imaging provides a promising pathway for far-field, non-invasive, pure-optical super-resolution. Shrinking the central focal spot is a common way used in SO design to pursue finer resolution. However, this inevitably results in huge side-lobes and low central intensity. This results in a severely restricted field of view (FOV) and poor image quality, hindering practical single-shot imaging. Here, we propose an end-to-end joint optimization framework combining the physical SO light field with a backend computational method. Instead of optimizing the isolated SO spot, we directly incorporate the ultimate resolving capability into the optical design loop. Specifically, the physical point spread function is parameterized and optimized by simulating the entire forward imaging and deconvolution process, effectively avoiding the severe noise amplification and reconstruction artifacts that typically plague conventional super-oscillatory field designs, and enhancing the resolution. As a demonstration, our system successfully breaks the diffraction limit, resolving double-hole structures at 0.77 times the diffraction-limited resolution. Furthermore, as our deconvolution model is completely general and assumes no prior knowledge of the object, this scan-free approach is effectively extended to the super-resolution imaging of extended targets, successfully overcoming conventional FOV limitations.