Fig. 4. Evaluation on the noise robustness of the proposed algorithm.

Near-field propagation images as would be recorded by the camera in Fig. 1A (top row) and reconstructed images (middle and bottom rows) in the presence of simulated Poisson noise with a total number of photons n_ph used in data recording. The middle row projection images correspond to the second column image of Fig. 2B, while the bottom row slice images correspond to the first column image of Fig. 2B. As can be seen, reducing the number of photons n_ph within the object (and, thus, the incident number of photons per pixel per viewing angle N_pix,θ as given by Eq. 8) leads to a decrease in image fidelity and signal-to-noise ratio (SNR) at the single-pixel level. One can also evaluate this as a loss of spatial resolution with decreasing exposure, as shown in Fig. 5.