Assessment of advanced windowing techniques for digital particle image velocimetry (DPIV)

The Fourier-based cross-correlation is the most common evaluation technique for DPIV estimation, due to its computational simplicity. However, because Fourier transforms are taken over discrete finite size regions, systematic errors are introduced due to the improper filtering of the input signals. This study explores the potential of advanced windowing techniques to attenuate these Fourier-based errors. The choice of window is shown to impact the spatial resolution, the measurement accuracy and the peak detection process. Error analysis using artificial image simulations is able to characterize a set of optimal windows onto a single performance characteristic. Using this analysis, a set of criteria is defined for an optimal windowing from which the analysis focused on the use of the 50% Gaussian window. The Gaussian window is further compared against standard evaluation techniques in both shear and vortex simulations, which indicate substantial performance benefits with this advanced technique. Further simulations reveal that background noise greatly amplifies the loss of correlation errors, which affect the peak detection process. However, these effects are easily overcome through the use of image preprocessing or the robust phase correlation. Images from the 2003 PIV challenge are used to validate the Gaussian window technique, which is able to remove nearly all of the erroneous vectors in comparison to standard windows.