Robust Imaging Against Laser Dazzling via Cross-Spectral Information Complementation

High-intensity laser illumination of an imaging system can produce complex spot occlusion and interference in the captured images, resulting in partial loss of scene information. To enable effective detection in complex laser environments, we propose an optics–algorithm co-designed imaging method for laser-interference removal. A four-aperture, band-partitioned imaging architecture is employed to acquire multichannel scene information. Laser dazzling causes severe interference only in the channel corresponding to its wavelength, while the remaining three channels preserve the information necessary for scene inference and reconstruction. We further develop a laser-glare image reconstruction algorithm based on cross-spectral information complementation, termed the Cross-Aperture Spectral-Prior Restoration Network (CASPR-Net). Based on the severity of laser interference, we designed two restoration branches for strong and weak interference, addressing saturation-induced information loss and halo- and speckle-induced radiometric and texture deviations, respectively. Subsequently, a multi-scale global branch was introduced to enhance radiometric consistency and structural details in the restored results. Laser-glare images synthesized from public spectral datasets and real captured laser spots are used for network training, and a four-aperture spectral imaging system is constructed to capture real laser-glare images for experimental validation. The experimental results demonstrate that the proposed optics–algorithm co-designed method can effectively reconstruct images degraded by laser interference. This work provides a feasible approach to robust imaging under complex illumination conditions and has potential applications in autonomous driving, security surveillance, and related fields.