Single-shot coded aperture polarization imaging system with fast momentum primal-dual hybrid gradient algorithm

Conventional polarization imaging systems employing optical lenses encounter fundamental constraints in miniaturization due to the inherent bulkiness of lens components. We propose a computational polarization imaging system that synergistically combines a random mask with a division-of-focal-plane (DoFP) polarization sensor. The architecture enables single-exposure acquisition of four-channel polarization states while maintaining compact configuration and cost efficiency. The light rays emitted from each point on the object plane are encoded by the random mask, forming overlapped intensity patterns on the image plane that require computational reconstruction. To address the inverse problem, we develop a momentum-accelerated primal-dual hybrid gradient algorithm (M-PDHG) that incorporates the historical gradient momentum for convergence acceleration, a dual-variable iterative framework for computational efficiency, and adaptive regularization for noise suppression. Experimental validation confirms the system's capability in polarization imaging with reduced noise amplification, demonstrating its practical potential for lightweight polarization imaging applications.