Non-invertible and revocable iris templates using key-dependent wavelet transforms

Abstract

A technique to generate non-invertible and revocable iris templates is proposed employing key-dependent wavelet transforms. In particular, parametrised wavelet-filters and wavelet packets are used in feature extraction in replacement of a pyramidal D4 wavelet transform. Since the template generation process is non-invertible by design, the overall scheme is non-invertible as well. Recognition accuracy is found to be high as long as personal tokens remain secret, templates can be revoked by simply exchanging the wavelet transform applied in the feature extraction process.