Sensitivity of image-preprocessing filters in LSPIV/STIV: impacts on velocity accuracy from laboratory to river conditions

Large-scale particle image velocimetry (LSPIV) and space-time image velocimetry (STIV) are increasingly applied for flow velocity estimation in both laboratory and field environments. However, their measurement accuracy is highly sensitive to image preprocessing, particularly filtering operations. This study systematically investigates the sensitivity of five representative filtering methods—spatial-domain, frequency-domain, adaptive, morphological, and multiscale/wavelet filtering—to parameter variations and their impacts on velocity estimation accuracy. Using “typical flow field images” from controlled laboratory experiments and natural river environments as benchmarks, we compare the trade-offs between noise suppression and feature preservation across diverse scenarios. Results demonstrate that filtering parameters exhibit a two-sided constraint: excessive filtering leads to signal attenuation and underestimated velocities, whereas insufficient filtering allows residual noise to destabilize correlation matching. Optimal accuracy is achieved within a balance zone where particle features are preserved while background noise is effectively suppressed. Spatial-domain and adaptive filters are generally effective for laboratory-scale, low-turbulence conditions, while frequency-domain and wavelet-based approaches show greater robustness in large-scale, high-turbulence rivers. These findings provide practical guidance for tailoring preprocessing strategies to specific hydrodynamic contexts and highlight the essential role of filtering in improving the stability and reliability of image-based flow measurements.