Graph matching iris image blocks with local binary pattern

TL;DR: In this article, the histogram of local binary pattern was used for global iris texture representation and graph matching for structural classification to complement the state-of-the-art methods with orthogonal features and classifier.

Abstract: Iris-based personal identification has attracted much attention in recent years. Almost all the state-of-the-art iris recognition algorithms are based on statistical classifier and local image features, which are noise sensitive and hardly to deliver perfect recognition performance. In this paper, we propose a novel iris recognition method, using the histogram of local binary pattern for global iris texture representation and graph matching for structural classification. The objective of our idea is to complement the state-of-the-art methods with orthogonal features and classifier. In the texture-rich iris image database UPOL, our method achieves higher discriminability than state-of-the-art approaches. But our algorithm does not perform well in the CASIA database whose images are less textured. Then the value of our work is demonstrated by providing complementary information to the state-of-the-art iris recognition systems. After simple fusion with our method, the equal error rate of Daugman's algorithm could be halved.