Enhancing iris recognition framework using feature selection and BPNN

Iris recognition has an important role in the field of biometrics since mid twentieth century. Among all biometric characteristics, iris is well-known to own a rich, stable and unique set of features. Different features like phase based features, textural features, key point descriptors and zero-crossing representations have been used to carry out iris recognition in the past. In this paper, two prevailing sets of features are combined to be used for iris recognition: first-order and second-order statistical measures as textural feature descriptors. A hybrid statistical dependency based feature selection algorithm is also applied on the extracted feature descriptors to remove noisy and redundant feature, thus reducing the size of feature vector. Back propagation neural network using Levenberg-Marquardt training algorithm is used for recognition task. The proposed iris recognition framework is tested on a well-known iris datasets like CASIA V1, CASIA V3: interval and lamp, and UBIRIS V1 to evaluate the robustness, and showed promising outcomes with the best genuine acceptance rate of 99.38, 99.84, 98.24 and 96.24% for a feature vector of size 13–17.