Photon-limited imaging of weak scattering objects is strongly affected by shot noise, leading to reduced contrast and variable phase retrieval. Introduction of phase-encoding elements can improve image formation by converting phase variations into measurable intensity modulations; however, conventional phase plates such as Zernike or vortex designs impose fixed symmetries and often lack robustness when photon budgets are low. This work dives into a simulation-based investigation of random binary phase plates as an alternative encoding approach for photon-limited computational imaging.
Image formation is modelled using scalar diffraction theory and angular-spectrum propagation through transmission phase plates composed of randomly distributed binary phase regions. Low dose detection is explicitly incorporated using Poisson noise statistics. Image recovery is performed using regularized inverse filtering, followed through systematic comparison of different phase-encoding methods devoid of complex reconstruction algorithms.
The results indicate that random binary phase encoding introduces wavefront diversity that possibly improves spatial–frequency mixing and stabilizes image reconstruction under strong shot noise. Relative to conventional symmetric phase plates, the random designs exhibit better CNR [Contrast to noise ratio] and reduced sensitivity to photon loss at detected photon levels below 10³. The dependence of imaging performance on phase-plate feature size is also examined, identifying parameter combinations that enhance noise robustness while maintaining reconstruction fidelity.