Abstract
In recent years, cross-spectral iris recognition has emerged as a promisingbiometric approach to establish the identity of individuals. However, matchingiris images acquired at different spectral bands (i.e., matching a visible(VIS) iris probe to a gallery of near-infrared (NIR) iris images or vice versa)shows a significant performance degradation when compared to intraband NIRmatching. Hence, in this paper, we have investigated a range of deepconvolutional generative adversarial network (DCGAN) architectures to furtherimprove the accuracy of cross-spectral iris recognition methods. Moreover,unlike the existing works in the literature, we introduce a resolutiondifference into the classical cross-spectral matching problem domain. We havedeveloped two different techniques using the conditional generative adversarialnetwork (cGAN) as a backbone architecture for cross-spectral iris matching. Inthe first approach, we simultaneously address the cross-resolution andcross-spectral matching problem by training a cGAN that jointly translatescross-resolution as well as cross-spectral tasks to the same resolution andwithin the same spectrum. In the second approach, we design a coupledgenerative adversarial network (cpGAN) architecture consisting of a pair ofcGAN modules that project the VIS and NIR iris images into a low-dimensionalembedding domain to ensure maximum pairwise similarity between the featurevectors from the two iris modalities of the same subject.