Real-time minimalist imaging via spatial aberration uniformization and re-parameterized restoration

Minimalist optical systems typically suffer from spatial aberration non-uniformity across different fields of view (FoVs), degrading restoration quality, while the high complexity of existing networks limits real-time efficiency. This paper presents a comprehensive design framework that encompasses a spatial aberration uniformization (SAU) methodology and a lightweight real-time re-parameterized image restoration network (RPIRNet). By integrating a spatial aberration uniformity metric into the optical design process, imaging consistency across the full FoV is substantially improved. In the subsequent restoration stage, we adopt a structural re-parameterization strategy to simplify the network architecture during inference without sacrificing performance. Additionally, a frequency-aware loss function is introduced to enhance the recovery of high-frequency components. To validate the proposed framework, a dual-lens optical system is designed and fabricated, achieving high-quality imaging with a real-time inference latency of 2.15 ms. These results demonstrate the feasibility and strong potential of our approach for practical applications.