This work explores the possibility of mixing two different fingerprints, pertaining to two different fingers, at the image level in order to generate a new fingerprint. To mix two fingerprints, each fingerprint pattern is decomposed into two different components, viz., the continuous and spiral components. After prealigning the components of each fingerprint, the continuous component of one fingerprint is combined with the spiral component of the other fingerprint. Experiments on the West Virginia University (WVU) and FVC2002 datasets show that mixing fingerprints has several benefits: (a) it can be used to generate virtual identities from two different fingers; (b) it can be used to obscure the information present in an individual's fingerprint image prior to storing it in a central database; and (c) it can be used to generate a cancelable fingerprint template, i.e., the template can be reset if the mixed fingerprint is compromised.

Preserving the privacy of digital biometric data (e.g., face images) stored in a central database has become of paramount importance. This work explores the possibility of using visual cryptography for imparting privacy to biometric data such as fingerprint images, iris codes, and face images. In the case of faces, a private face image is dithered into two host face images (known as sheets) that are stored in two separate database servers such that the private image can be revealed only when both sheets are simultaneously available; at the same time, the individual sheet images do not reveal the identity of the private image. A series of experiments on the XM2VTS and IMM face databases confirm the following: 1) the possibility of hiding a private face image in two host face images; 2) the successful matching of face images reconstructed from the sheets; 3) the inability of sheets to reveal the identity of the private face image; 4) using different pairs of host images to encrypt different samples of the same private face; and 5) the difficulty of cross-database matching for determining identities. A similar process is used to de-identify fingerprint images and iris codes prior to storing them in a central database.