In this paper we propose EyeCerts, a biometric system for identification of people which achieves off-line verification of certified, cryptographically secure documents. An EyeCert is a printed document which certifies the association of a given text with a biometric feature - a compressed version of a human iris in this work. As a central component of the yeCert system, we present an iris analysis technique that extracts and compresses the unique features of a given iris using limited storage. The compressed features should be at maximal distance with respect to a reference iris image database. The iris analysis algorithm performs several steps in three main phases: (i) it detects the human iris, (ii) it converts the isolated iris using a modified Fourier-Mellin transform into a standard domain where the common radial patterns of the human iris are concisely represented, and (iii) it optimally selects, aligns, and near-optimally compresses the most distinctive transform coefficients for each individual user. Using a low quality imaging system (sub-US$100) and developed and readily available low complexity processing techniques, the overall system is shown to have probabilities of false negative and false positive on the order of 10^-5.