Optical attention mechanism for high-resolution computational imaging

Traditional optical design (TOD) achieves high-resolution imaging through complex stacks of lenses. Simplifying the lens while ensuring high-resolution capability has been a difficult challenge. In TOD, high resolution means small spot size, so all lens surface areas (LSAs) are constrained to synergistically converge the rays. Here, we mimic the selective attention ability in the human cognitive system and propose the optical attention (OA) mechanism. Compared to TOD, OA selectively attends on LSAs that are conducive to high-frequency information transmission (HFIT), rather than treating all LSAs equally. OA can fully utilize the HFIT capability of simple lens by dynamically distributing local regularization on LSAs, and then the influence of residual aberrations is removed by image restoration, thereby achieving high-resolution computational imaging. Simulation and real-world experiments demonstrate that OA has a wide range of applicable scenarios. On the one hand, it can simplify the complex lens, reducing the number of lenses by 33% and the total track length (TTL) by 18.5% while ensuring the same high-resolution imaging performance. On the other hand, it can improve the imaging performance of the single lens. Compared with existing computational imaging methods, it can increase the contrast of high spatial frequencies (100 lp/mm) by more than 55%. Therefore, the proposed OA significantly simplifies the lens while ensuring high-resolution capability, which expands and diversifies the optical design methods and paves the way for high-performance imaging with lighter and smaller devices.