Optimize the ordering of the Hadamard basis through the max-projection method to design the masks of ghost imaging suitable for efficient reconstruction of face images

One important component of ghost imaging is the encoded mask. Hadamard matrix has been the widely used mask for its binary and orthogonal property. Inspired by deep learning, max-projection ordering was proposed to reorder Hadamard bases for efficient and fast reconstruction applied to frontier research such as cerebral imaging. Simulations of face demonstrated only 20 training images obtained an outstanding ordering. In noise-free simulation at an ultra-low sampling rate of 5%, PSNR of max-projection ordering was 1.1dB higher than those of cake-cutting ordering with the best performance in the reference group. Reconstruction time was reduced to milliseconds. In noisy simulation, at ultra-low sampling rates, the retrieved images were almost identical to those without noise. Accordingly, max-projection ordering Hadamard matrix is a promising solution to real-time ghost imaging for higher reconstruction quality, stronger noise immunity and millisecond reconstruction time.