INL-InSAR: An Improved Nonlocal Means Algorithm for InSAR Phase Filtering

Objectives: Phase filtering is a critical step in interferometric Synthetic Aperture Radar (InSAR) data processing, and the quality of the interferometric phase directly affects the accuracy of subsequent information extraction. Traditional nonlocal methods rely on adjusting the search window to increase the number of similar pixels. However, this approach increases the computational complexity of the algorithm and provides limited similar-pixel improvement in regions with weak structural similarity. Furthermore, in existing methods, similar-pixel selection and filtering parameters generally lack joint optimization, neglecting the impact of their combined optimization on filtering performance. Pre-filtered interferograms can typically provide a relatively accurate reference for pixel similarity computation. Nevertheless, when the pre-filtered image suffers from significant phase information loss, substantial phase estimation errors may occur. Methods: To address these issues, an improved nonlocal means InSAR phase filtering (INL-InSAR) algorithm is proposed. Firstly, noisy interferograms covering the study area are obtained using InSAR technology. Then, a fusion strategy based on phase offset-compensation (OC) and adaptive smoothing parameters is introduced into the algorithm. By calculating the constant offset-compensation value between two similar windows and searching adaptive filtering coefficients with a defined step size, this strategy effectively increases the number of similar pixels and addresses issues of over-smoothing or under-smoothing in local regions. Finally, an iterative filtering framework is established, where each iteration generates a more reliable reference image, significantly improving the accuracy of similarity computation and avoiding the limitations of using a single pre-filtered image. Results: In the simulation experiments, the proposed method was able to recover clearer interference fringe patterns and exhibited better phase continuity. In the real-data experiments, the residue removal rates reached 99.87% and 99.61% and the Phase Standard Deviation (PSD) was reduced by 3.87% and 11.74% lower than the suboptimal method, respectively. In addition, the influence of filtering window, the initial filtering parameters, and the iterative process on the filtering performance were also evaluated. The experimental results show that the parameters selected for the INL-InSAR method are optimal, and the proposed method achieves the best filtering performance through the fusion strategy and iterative framework. Conclusions: The proposed method can effectively suppress phase noise and better preserve phase detail information, enabling higher-quality subsequent information extraction.