Robust holographic imaging for real-world applications with joint optimization

Abstract

<p>Digital inline holography offers a compact, lensless imaging solution, but its practical deployment is often hindered by the need for precise system alignment and calibration, particularly regarding propagation distance. This work presents J-Net, a robust, untrained neural network that significantly mitigates these limitations. J-Net eliminates the need for prior knowledge or calibration of the propagation distance by simultaneously reconstructing both the complex-valued object magnitude and the propagation distance from a single hologram. This inherent robustness to distance variations makes J-Net highly practical for real-world applications where precise system control is difficult or impossible. Experimental results demonstrate high-quality amplitude and phase reconstruction even under mismatched distance conditions, showcasing J-Net’s potential to enable robust deployment of holographic imaging across diverse fields.</p>