Poor-Contrast Particle Image Processing in Microscale Mixing

Particle image velocimetry (PIV) often employs the cross-correlation function to identify average particle displacement in an interrogation window. The quality of correlation peak has a strong dependence on the signal-to-noise ratio (SNR), or contrast of the particle images. In fact, variable-contrast particle images are not uncommon in the PIV community: Strong light sheet intensity variations, wall reflections, multiple scattering in densely-seeded regions and two-phase flow applications are likely sources of local contrast variations. In this paper, we choose an image pair obtained in a micro-scale mixing experiment with severe local contrast gradients. In regions where image contrast is sufficiently poor, the noise peaks cast a shadow on the true correlation peak, producing erroneous velocity vectors. This work aims to demonstrate that two image pre-processing techniques — local contrast normalization and Difference of Gaussian (DoG) filter — improve the correlation results significantly in poor-contrast regions.