Coded aperture compressive temporal imaging using complementary codes and untrained neural networks for high-quality reconstruction

The coded aperture compressive temporal imaging (CACTI) modality is capable of capturing dynamic scenes with only a single-shot of a 2D detector. In this Letter, we present a specifically designed CACTI system to boost the reconstruction quality. Our design is twofold: for the optical encoder, we use complementary codes instead of random ones as widely adopted before; for the reconstruction algorithm, an untrained neural network-based algorithm is developed. Experimental and simulation tests show that such co-design of encoding-decoding produces superior image quality over other CACTI schemes using random codes and other optimization algorithms. In addition, a dual-prism design in the optical system improves the light efficiency by approximately a factor of four compared with previous systems.