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

摘要: 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 high (HR) gallery samples through pixel-level or featurelevel super-resolution. We argue that binary codes feature templates are more directly relevant recognition performance. This paper proposes code-level scheme for heterogeneous matching and HR images. statistical relationship between number code corresponding latent image is established based on an adapted Markov network. Moreover, cooccurence neighboring bits also modeled this So we can obtain enhanced from probe set sequences. In addition, weight mask be derived model, used further accuracy. Experimental results Quality-Face/Iris Research Ensemble (Q-FIRE) database demonstrate information fusion performs significantly better than pixel-level, feature-level score-level