Toward practical remote iris recognition: A boosting based framework

Abstract In this paper, we present a generalized boosting framework to tackle some challenging problems in practical remote iris recognition, namely, iris detection, iris mislocalization detection, iris spoof detection as well as iris recognition. This solution takes advantages of a set of carefully designed features and well-tuned boosting algorithms. Basically, there are two major contributions. The first one is an exploration into the intrinsic properties of remote iris recognition as well as the carefully designed robust features for specific problems. For example, the randomness of iris texture is explored, and ordinal measures are adopted as features for iris representation. The second major contribution is the methodology on how to tune Adaboost learning for specific problems. For instance, an effective similarity oriented boosting algorithm is proposed for iris recognition inspired by the similarity property of the training samples. Other specific contributions include: an efficient topological model of Haar-like features for robust iris detection, a texture and Adaboost based method for efficient iris spoof detection and iris mislocalization detection, a novel Gaussian model for adaptive decision making, etc. Extensive experiments on challenging iris image databases are conducted to evaluate the usefulness of the proposed methods, and the results show that state-of-the-art performance is achieved.