Benchmark of deep learning approaches for phase denoising in digital holography

Abstract

This paper presents a comparative study of deep learning based algorithms to de-noise wrapped phase maps in digital holography interferometry. In order to compare two deep neural networks trained on two different databases, we propose to train both networks on both databases. The four resulting networks are then benchmarked with one unique database. We present the assessment between two models developed in Python. A third model developed in matlab is iadded in evaluation presented in this paper but will be not subject to retraining in the second step of the benchmark.