Detection of Contact-Lens-Based Iris Biometric Spoofs Using Stereo Imaging

Cosmetic contact lenses can be used to spoof an iris biometric system, either to evade being matched to a watch list or in principle even to masquerade as a selected other person. Existing approaches to detecting whether or not a person is wearing cosmetic contact lenses either are limited to detecting lenses created by a particular manufacturing technology, assume knowledge of the particular pattern printed in/on the lens, or require a sequence of images. We present proof-of-concept results for a method of detecting cosmetic contact lenses that is general, in the sense that it assumes nothing about the manufacturing technique or texture pattern of the lens, and that requires only a "snapshot" instance of imaging. The "snapshot" is a stereo pair of images, from which the shape of the surface of the iris texture region is estimated. In the absence of contacts or the presence of clear contacts, the iris region presents a coarse planar surface. In the presence of cosmetic contacts, the iris region presents a convex surface. Thus the problem of determining if a person is wearing a cosmetic contact lens is transformed into the problem of classifying the estimated surface shape for the iris region. This is the first approach to analyze iris biometric images in the context of 3D shape.