Fourier ptychographic layer-based imaging of hazy environments

Abstract

Macroscopic Fourier ptychographic imaging method can significantly improve the resolution of imaging systems. Nevertheless, under far-field complex environment conditions, such as strong scattering medium (e.g.,hazy), the inhomogeneous distribution of the particles inside the hazy cause random interference to the propagation of the photons carrying the target information, resulting in distortion and blurring of the wavefront information. This severely impacts the quality of the final reconstructed image. In this paper, we propose an improved Resnet network to extract and enhance the features from distorted intensity images in hazy environment, and effectively merge information through multi-scale feature fusion. We successfully recover the target intensity information sequences hidden by different concentrations of hazy and even dynamic non-uniform hazy, and present the final Fourier ptychographic reconstruction results. To the best of our knowledge, we are the first to demonstrate the feasibility of far-field Fourier ptychographic imaging through hazy environments using deep learning methods. Experimental results indicate that it can effectively overcome the images blurring and distortion problems caused by the scattering of hazy environment, significantly improving the structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR) of final macroscopic Fourier ptychographic reconstructed images under different hazy levels. This study not only reveals the great value of deep learning in solving the imaging problems of complex scenes such as dynamic non-uniform strong scattering media, but also effectively suppresses the adverse effect of hazy scattering environment on macroscopic Fourier ptychographic imaging, which greatly promotes the progress and development of the application of the Fourier ptychographic imaging technology in the field of imaging complex environments in the long distance.