Metasurface-based continuous nonuniform sampling for demoiréing

Abstract

Moiré patterns, which are irregular interference stripes, arise from the interference between high-frequency scenes and the spatial discretization of imaging cameras. Typical optical demoiréing methods employ low-pass filters to remove the high-frequency components responsible for generating moiré patterns. However, these methods will reduce image detail. The arrangement of photosensitive pixel arrays is confined to a regular pattern in semiconductor fabrication. This limitation imposes severe restrictions on related optical designs. In this work, we propose a metasurface-based random microlens array (RMLA) to fundamentally eliminate moiré artifacts. The proposed architecture features continuous and randomly distributed light-collection regions coupled with regularly arranged focal points aligned to the pixel grid. This configuration imposes a physically nonuniform light-footprint on the CMOS pixels, effectively decoupling the sampling process from the scene periodicity. Consequently, spectral aliasing in the luminance channel is intrinsically suppressed, theoretically approximating continuous nonuniform sampling. Both simulation and experimental results verify that moiré patterns induced by signals near the Nyquist limit are effectively eliminated without sacrificing spatial resolution. Unlike computational approaches requiring heavy post-processing, our method offers a passive optical solution that can be optically coupled with commercial off-the-shelf CMOS sensors without modifying the underlying circuitry. This work opens up a new dimension in optical design, offering a viable path to alias-free imaging systems.