Deep learning assisted image transmission in multimode fibers

We propose a data -driven approach for light transmission control inside multimode fibers (MMFs). Specifically, we show that a convolutional neural network is able to reconstruct amplitude/phase modulated images from scrambled amplitude -only images obtained at the output of a 0.75m long MMF with a fidelity (correlation) as high as 98%. We show that the trained network shows good generalization as well. In particular, it is shown that the network is able to reconstruct images that do not belong to train/test datasets.