Quasi noise-free reconstruction of long-wavelength digital holograms

Digital Holography (DH) suffers from severe degradation of the reconstruction quality due to the presence of speckles. Speckle is due to the source coherence and shows on the hologram as a multiplicative, correlated noise. Due to the larger size of the speckle grains, the lower resolution, and the worse features of the available hardware, long wavelength digital holography is more severely degraded by noise than its visible wavelength counterpart is. Non-Bayesian approaches to the denoising problem suffer from resolution loss or complex acquisition systems required to record multiple uncorrelated holograms to be averaged. Instead of providing multiple captures, these can be simulated to yield a number of reconstructions from one single hologram (generally referred to as numerical Multi-Look, ML). However, the ML improvement is inherently bounded to a theoretical limit. On the other hand, image processing has offered a wide literature on the topic over the last decades. Among the most efficient methods to reduce additive Gaussian noise, 3D Block Matching (BM3D) has emerged and it is nowadays widely used in the image processing framework. However, BM3D performance worsens in the presence of speckle and cannot be effectively applied to long wavelength DH. Here we show that the joint action of numerical ML (thought as a preprocessing filter) and BM3D in post-processing permits to overcome the theoretical limit of ML and to outperform the BM3D for the denoising of holograms. The quasi noise free reconstruction of long wavelength holograms of famous artworks will be shown.