Improved SNR and super-resolution reconstruction of multi-scale digital holography based on deep learning

Abstract

Digital holography is one of the key technologies to obtain the wavefront information of three-dimensional objects, and obtaining high quality hologram is the primary condition. Due to the constraints of image sensors and the influence of experimental environment, the digital hologram obtained has the problems of speckle noise and low resolution. In order to overcome this constraint, a method based on deep learning is used to improve the hologram quality and fringe signal-to-noise ratio, and then high-quality reconstructed phase diagrams are obtained as data sets to realize end-to-end superresolution reconstruction. The results show that the obtained high quality hologram has better reconstruction effect, and the effect of speckle noise is suppressed. The performance of three loss functions in the training of the network is compared.