Thin and lightweight camera based on Pancake lens and deep learning

Abstract

Computational imaging using a Pancake lens can help reduce the size of optical systems by folded optics. However, Pancake cameras frequently exhibit inferior image quality due to stray light, low light transmission, and spatially varying aberrations. In this Letter, we propose a thin and lightweight camera comprising a polarization-based catadioptric Pancake lens and a Fourier Position encoding Network (FPNet). The camera achieves high-quality imaging at an f-number of 0.4 and an expansive 88° field of view. The FPNet encodes the positional order of the point spread functions, mitigating global optical image degradation and improving image quality by 10.13 dB in PSNR. The Pancake camera and FPNet have potential applications in mobile photography and virtual/augmented reality.