Proper orthogonal decomposition based outlier correction for PIV data

Particle image velocimetry (PIV) is a powerful tool to study complex flows quantitatively. Post-processing of PIV data is necessary for outlier correction (OC) because of the image noise. Traditional methods detect and correct spurious vectors, respectively, using local statistical models. A new method proposed in this paper iteratively detects and replaces outliers using proper orthogonal decomposition (POD), which can dynamically approximate the original pure velocity field. The new algorithm, named as POD-OC, reconstructs a reference velocity field using low-order POD modes to detect outliers and uses that reference field for OC as well. Compared with the method of normalized median test, POD-OC is more efficient for detecting clustered outliers. It is also more accurate than other common interpolation approaches on outlier fixing. A novel block POD-OC is also designed for post-processing on an instantaneous velocity field, which overcomes the limit that POD can only be applied on a dataset with a large number of instantaneous fields.