Nonlinear holographic image inpainting via deep learning

Nonlinear photonic crystals can be employed to generate holographic images in nonlinear optical processes. However, due to the limit of binary structure, the process of holographic imaging will lose part of the amplitude information and cause image hollowing, thus the imaging quality is reduced. Generative adversarial network, a deep learning network based on game theory, is used to restore images. The results of restoration show high similarity to the original images, effectively weakening the effect of image hollowing, suppressing the diffraction effect, and restoring grayscale values. This image post-processing approach completes the field of application of nonlinear holographic imaging, which is useful for non-visible source imaging, crosstalk avoidance, optical encryption, and so on.