Optical image hiding under low light illumination based on deep learning

Abstract

We propose a system that realizes optical image hiding under low light and reconstructs secret images through deep learning. Based on the improved Mach–Zehnder interferometer system, the light intensity of the two optical paths of the reference light and the object light is adjusted by the neutral density filter (NDF) until the light intensity ratio is 1:15, which constitutes the experimental system of optical image hiding under low light field in this article. The interference hologram collected by the CCD is used as the training data set of the neural network, and the neural network model is continuously optimized. Finally, the optimized neural network model reconstructs the secret image with high quality. The proposal of this scheme reduces the embedding depth of the secret image, making the secret image less noticeable. In addition, the results of optical experiments and simulated test results of different light intensity ratios show that the scheme has certain feasibility and effectiveness.