Reduced CNNs architectures applied to phase maps corrupted with speckle noise

This paper addresses the problem of phase images corrupted with speckle noise. DnCNN residual networks with different depths were built and trained with various holographic noisy phase data. All models are evaluated in terms of phase error with HOLODEEP benchmark data and with 3 unseen images corresponding to different experimental conditions. The best results are obtained using a network with only 4 residual blocks, and trained with a wide range of noisy speckle patterns.