Multiscale principal contour direction for varying lighting face recognition

Proposed is a new type of image feature for varying lighting face recognition, namely, multiscale principal contour direction (MPCD). It can capture `true` neighbourhood contour direction information and is robust to different illumination conditions. In addition, MPCD is constructed for extracting illumination invariant by applying the non-subsampled contourlet transform (NSCT) to multiscale contour analysis. To evaluate its performance, a series of experiments are performed on two well-known illumination databases (i.e. Yale B and CMU PIE), showing the efficacy of the method.