Fringe pattern preprocessing via adaptive multidirectional empirical mode decomposition

Fringe patterns often suffer from background illumination and noise due to the environment changes, the sample itself and the optical setup. Therefore, in the preprocessing stage before phase retrieval, removing the background, and minimizing noise is crucial for the accuracy of optical measurements. In this contribution, we propose an automatic, robust, and effective fringe pattern preprocessing based on adaptive multidirectional empirical mode decomposition (AMDEMD). AMDEMD utilizes customized directional fitting filters based on the local direction of the data to address the deficiency in local direction constraints. Furthermore, by examining the decomposition outcomes, AMDEMD introduces what we believe to be novel inner and outer stopping criteria for sifting to automatically extract the intrinsic mode functions containing fringes and the residual component containing the background. This greatly simplifies the reconstruction process. Before decomposition, the adaptive noise level estimation based block matching 3D filtering is applied to the fringe pattern to ensure robustness in noise removal. Performance validation of the previously reported bidimensional empirical mode decomposition is conducted by simulated and experimental data to verify the versatility and effectiveness of the proposed method.