Race recognition from face images using Weber local descriptor

This paper proposes a race recognition system from face images based on Weber local descriptors (WLD). In the system, first, WLD histogram is extracted from normalized face images. Then Kruskal-Wallis feature selection technique is used to select the best discriminated bins. City block, Euclidean, and chi-square minimum distance classifiers are used for testing. In the experiments, FERET database is used where there are five major race groups: Asian, African or American Black, Hispanic, Middle-Eastern, and White. Experimental results show that the proposed system with WLD histogram, feature selection, and city block distance classifier achieves accuracies of Asian: 97.74%, Black: 96.89%, Hispanic: 92.06%, Middle: 98.33%, and White: 99.53%. These accuracies are significantly higher than those using principal component analyses.