High-speed computational ghost imaging based on an auto-encoder network under low sampling rate.

Abstract

Computational ghost imaging is difficult to apply under low sampling rate. We propose high-speed computational ghost imaging based on an auto-encoder network to reconstruct images with high quality under low sampling rate. The auto-encoder convolutional neural network is designed, and the object images can be reconstructed accurately without labeled images. Experimental results show that our method can greatly improve the peak signal-to-noise ratio and structural similarity of the test samples, which are up to 18 and 0.7, respectively, under low sampling rate. Our method only needs 1/10 of traditional deep learning samples to achieve fast and high-quality image reconstruction, and the network also has a certain generalization to the gray-scale images.