Denoising methods for time-resolved PIV measurements

The increasing capabilities of currently available high-speed cameras present several new opportunities for particle image velocimetry (PIV). In particular, temporal postprocessing methods can be used to remove spurious vectors but can also be applied to remove inherent noise. This paper explores this second possibility by estimating the error introduced by several denoising methods on manufactured velocity fields. It is found that PIV noise, while autocorrelated in space, is uncorrelated in time, which leads to a significant improvement in the efficiency of temporal denoising methods compared to their spatial counterparts. Among them, the optimal Wiener filter presents better results than convolution- or wavelet-based filters and has the valuable advantage that no adjustments are required, unlike other methods which generally involve the tuning of some parameters that depend on flow and measurement conditions and are not known a priori. Further refinements show that denoised data can be successfully deconvolved to increase the accuracy of remaining small-scale velocity fluctuations, leading in particular to the recovery of the true shape of turbulent spectra. In practice, the computation of the filter function is not always accurate and different procedures can be used to improve the method depending on the flow considered. Some of them are derived from the properties of the time-frequency spectrum provided by the wavelet transform.