Code-level information fusion of low-resolution iris image sequences for personal identification at a distance

Iris images captured at a distance usually have low resolution (LR) iris texture regions, which may lose some detailed identity information. The existed approaches try to improve the similarity of these LR iris images to high resolution (HR) gallery samples through pixel-level or featurelevel super-resolution. We argue that binary codes of iris feature templates are more directly relevant to iris recognition performance. This paper proposes a code-level scheme for heterogeneous matching of LR and HR iris images. The statistical relationship between a number of binary codes of LR iris images and a binary code corresponding to the latent HR iris image is established based on an adapted Markov network. Moreover, the cooccurence relationship between neighboring bits of HR iris code is also modeled through this Markov network. So that we can obtain an enhanced iris feature code from the probe set of LR iris image sequences. In addition, a weight mask can also be derived from the Markov model, which can be used to further improve iris recognition accuracy. Experimental results on Quality-Face/Iris Research Ensemble (Q-FIRE) database demonstrate that code-level information fusion performs significantly better than existed pixel-level, feature-level and score-level approaches for recognition of low resolution iris image sequences.