Fig. 1. Block diagram of single-photon compression imaging system

Fig. 2. Influence of Poisson noise on the pixel value in a two-dimensional image.(a) Original pixel value; (b) actual pixel value affected by Poisson noise

Fig. 3. Effect of Poisson noise on the image under different measurement time, and the average photon count rate is 34672 s^-1. (a) Original image; (b) total sampling time is 20 s; (c) total sampling time is 50 s; (d) total sampling time is 1000 s

Fig. 4. Schematic of anti-noise network training method for single-photon compression imaging

Fig. 5. Diagram of DFC-net structure

Fig. 6. Diagram of RNP-net structure

Fig. 7. Schematic of RPN-net output

Fig. 8. Schematic of loss drop during RPN-net training

Fig. 9. Mask photos.(a) Physical image in a bright environment; (b) physical image illuminated in a dark environment

Fig. 10. PSNR of different training methods under different sampling time coefficients without background noise.(a) Sampling time factor is 60; (b) sampling time factor is 100; (c) sampling time factor is 140

Fig. 11. PSNR of different training methods under different background noises when the sampling time is constant. (a) Dark count rate is 5; (b) dark count rate is 10; (c) dark count rate is 15

Fig. 12. Rows represent the reconstructed results of different training methods and different reconstruction networks, the columns represent the reconstructed results at different sampling time (average number of photons η), and the size of all pictures is 64×64 when the sampling rate is a constant of 20%

Fig. 13. Average PSNR of natural pictures reconstructed by different reconstruction methods. (a) Sampling time factor is 60; (b) sampling time factor is 100; (c) sampling time factor is 140

Fig. 14. Rows represent the reconstructed results of different reconstruction methods, the columns represent the reconstructed results at different sampling time (average number of photons), and the size of all pictures is 64×64 when the sampling rate is a constant of 20%