Fringe processing using deep learning-assisted Fourier transform for a diffractive optical element-based background-oriented schlieren

This paper presents a noise-tolerant approach for processing the optical fringe patterns obtained in the diffractive optical element (DOE)-based background-oriented schlieren technique. The proposed method relies on reconstruction of normalized fringe patterns from images corrupted with severe noise using deep learning, and subsequent phase estimation via Fourier transform. The effectiveness of the proposed approach is validated through rigorous numerical simulations under varying levels of additive noise and speckle noise, as well as the presence of local fringe abnormalities. In addition, demonstration of the method’s utility on experimental fringe patterns is provided, thus showing the potential for quantitative flow visualization.