Noise-Robust Iris Authentication Using Local Higher-Order Moment Kernels

A novel biometric authentication method using kernel functions of higher-order statistical feature of the iris texture is introduced. When the observed iris images include noise, direct estimation and use of Gabor and local higher-order moment (LHOM) features for iris code generation suffers from performance degradation. In order to solve this issue, we propose to use the LHOM kernel function of pairs of local textures on a single iris image. In the experiments, the proposed method using LHOM kernels of orders 2 to 6 proved to be significantly robust against noise when compared with the conventional method.