Towards ethnicity detection using learning based classifiers

Soft biometric such as gender, age, and ethnicity detection grows quickly as biometric system detection to support other solid biometric modality such as face, fingerprint, and iris. Nowadays, ethnicity detection plays an important role for pre-identify and re-identity people with different demographic background. The goal of this paper is ethnicity detection from ocular biometrics. In this work, two different level of fusion have been applied. The first level of fusion is feature fusion which fused two important local features which are Local Binary Pattern (LBP), and Histogram of Oriented Gradient (HOG). The second level of fusion have been made at the derision level for the best results of the classifiers. Moreover, four classifiers which are support vector machine (SVM), Multi-Layer Perceptron (MLP), Linear Discriminant Analysis (LDA), and Quadratic Discriminant Analysis (QDA) with overall ten kernels have been trained, validate, and tested on the FERET database to show the performance before and after fusion. The best result with an overall test performance accuracy of 98.5% using SVM with polynomial kernel that have been applied on the fused features of LBP and HOG.