High-fidelity full-color correspondence imaging using single-pixel detectors in complex environments with random disturbances

Complex environments with random disturbances could induce dynamic scaling factors during single-pixel measurements, affecting reconstruction quality and color accuracy. In this Letter, we report high-fidelity full-color correspondence imaging (CI) in complex environments with random disturbances. A series of mixed illumination patterns, i.e., sinusoidal and random patterns, are designed and alternately projected onto an unknown target scene. At the receiving end, a trichroic prism is employed to guide optical waves into red, green, and blue channels, which are individually detected using a single-pixel detector. A series of dynamic scaling factors are estimated using the single-pixel light intensities collected when random patterns are used, which are further applied to correct the single-pixel light intensities collected when sinusoidal patterns are used. Then, a physics-enhanced neural network (PENet) without any training data is designed to perform the reconstruction. Finally, a color target scene can be retrieved by combining the reconstructions at three channels. A series of experiments are conducted to illustrate that the proposed method can achieve high-fidelity reconstruction with high accuracy of colors in complex scattering media with random disturbances. This study provides a promising solution for high-fidelity full-color imaging in complex environments with random disturbances.