Central difference information filtering phase unwrapping algorithm based on deep learning

Abstract

This paper presents a deep learning-based central difference information filtering phase unwrapping algorithm. First, the central difference information filter (CDIF) is introduced into phase unwrapping for wrapped phase images for the first time to our knowledge, and is combined with an efficient local phase gradient estimation technique, and an efficient path-following strategy based on heap sorting to construct a CDIF phase unwrapping procedure. Second, a region segmentation model based on deep learning is constructed to divide the wrapped phase images into a series of segmented regions along the edge of interferometric fringes, which is beneficial to improve the accuracy and efficiency of subsequent phase unwrapping. Finally, each segmented region of the wrapped phase images is unwrapped separately using the CDIF phase unwrapping procedure, and the unwrapping results of each segmented region are combined according to the phase consistency criterion of adjacent regions of the wrapped phase images to obtain the unwrapped phase of the whole wrapped phase images. Results delivered by the experiments on synthetic data and experimental data demonstrate the robustness and acceptable efficiency of the proposed method in phase unwrapping for the wrapped phase images, compared with some commonly used algorithms.