Improving iris recognition accuracy via cascaded classifiers

Abstract

As a reliable approach to human identification, iris recognition has received increasing attention in recent years. In the literature of iris recognition, local feature of image details has been verified as an efficient iris signature. But measurements from minutiae are easily affected by noises, which greatly limits the system's accuracy. When the matching score between two intra-class iris images is near the local feature based classifier's (LFC) decision boundary, the poor quality iris images are usually involved in matching. Then a novel iris blob matching algorithm is resorted to make the recognition decision which is more robust than the LFC in the noisy environment. The extensive experimental results demonstrate that the cascading scheme significantly outperforms individual classifier in terms of accuracy and robustness. © Springer-Verlag Berlin Heidelberg 2004.