Abstract
The paper introduces multiscale spatial Weber local descriptor (MSWLD) for robust face recognition system. In the proposed method, WLD is calculated in different neighborhood (multiscale) and WLD histograms are obtained from blocks of an image to preserve spatial information. WLD histograms from different blocks are then concatenated to produce the final feature set of a face image. Fisher ratio is applied to extract the dominant bins from the final WLD histogram. The MSWLD is evaluated on FERET and AT&T databases. In the experiments, the proposed method outperformed two state of the art techniques, namely, principal component analysis and local binary pattern.