Optical diffraction neural networks assisted computational ghost imaging through dynamic scattering media

TL;DR

The study tackles the challenge of imaging through dynamic scattering by integrating optical diffraction neural networks (ODNNs) into a computational ghost imaging framework. ODNNs, trained entirely on simulated data, actively correct scattering-induced distortions to preserve the correlation between illumination and reference patterns, enabling robust imaging through dynamic media. Experimental validation with rotating ground-glass diffusers demonstrates successful reconstructions across multiple coherence scales and shows the benefits of combining ODNN correction with untrained reconstruction methods under varying sampling conditions. The approach offers a fast, plug-and-play distortion-correction strategy with potential to extend to other imaging systems and modalities, while highlighting current limits in thicker or more complex scattering environments.

Abstract

Ghost imaging leverages a single-pixel detector with no spatial resolution to acquire object echo intensity signals, which are correlated with illumination patterns to reconstruct an image. This architecture inherently mitigates scattering interference between the object and the detector but sensitive to scattering between the light source and the object. To address this challenge, we propose an optical diffraction neural networks (ODNNs) assisted ghost imaging method for imaging through dynamic scattering media. In our scheme, a set of fixed ODNNs, trained on simulated datasets, is incorporated into the experimental optical path to actively correct random distortions induced by dynamic scattering media. Experimental validation using rotating single-layer and double-layer ground glass confirms the feasibility and effectiveness of our approach. Furthermore, our scheme can also be combined with physics-prior-based reconstruction algorithms, enabling high-quality imaging under undersampled conditions. This work demonstrates a novel strategy for imaging through dynamic scattering media, which can be extended to other imaging systems.

Figures (6)

• Figure 1: (a) Schematic of scattering-robust computational ghost imaging assisted by a two layers optical diffraction neural network. (b) Schematic diagram of ODNN training. First, the light field propagation from the DMD plane to the object plane is modeled. Subsequently, N different simulated objects and M different simulated scattering media are used as inputs to train a set of ODNNs. SM: Scattering Medium; DMD: Digital Micromirror Device.
• Figure 2: (a) Two groups of illumination patterns used for object image reconstruction. Group 1: patterns projected onto the DMD; Group 2: patterns of Group 1 after the forward physical process. (b) Schematic of image reconstruction performed by the untrained reconstruction algorithm.
• Figure 3: (a) Schematic diagram of the experimental setup for the ODNNs-assisted ghost imaging. PBS, Polarizing Beam Splitter; DMD, Digital Micromirror Device; SLM, Spatial Light Modulator. (b) A two-layer ODNN trained with a simulated dataset.
• Figure 4: Image reconstruction results obtained using illumination patterns with different transverse coherence scales (sizes). (a) Reconstruction result without ODNN correction. (b) and (c) show the reconstructed results based on the patterns from Group 1 and Group 2, respectively, when ODNN correction is implemented. (d) The iteration process curves of the untrained algorithm results in (b) and (c). (e) Quantitative comparison of the results in (a), (b), and (c), with PSNR as an example. (f) Reconstruction results of the untrained algorithm under 40 $\times$ 40 pixels illumination pattern at different numbers of iterations. (g) Speckle decorrelation curve of a rotating 220-grid ground glass.
• Figure 5: (a) Image reconstruction results of three objects at different sampling rates, where the illumination pattern size is 66×66 pixels. (b) and (c) show the PSNR and PCC metrics of the reconstruction results in (a), respectively. O1, O2, and O3 are abbreviations of object1, object2, and object3, respectively.