Ranklets: orientation selective non-parametric features applied to face detection

We introduce a family of multiscale, orientation-selective, non-parametric features ("ranklets") modelled on Haar wavelets. We clarify their relation to the Wilcoxon rank-sum test and the rank transform and provide an efficient scheme for computation based on the Mann-Whitney statistics. Finally, we show that ranklets outperform other rank features, Haar wavelets, SNoW and linear SVMs (based on independently published results) in face detection experiments over the 24\´045 test images in the MIT-CBCL database.