Fractal-Based Approach to Secure Key Generation from Fingerprint and Iris Biometrics

The generation of keys from users’ biometric data is gaining popularity in biometric-based cryptography systems. In such a system, managing keys and maintaining the security of biometric data are two challenging tasks. Existing methods, in general, use direct application of biometric templates, which can reveal the user’s biometric data and thus useless for future applications once it is compromised. To address this issue, researchers advocate multimodal biometrics combining two or more traits. This work aims generating a key from a person’s multimodal biometric traits with two modes: fingerprint, and iris. The work proposes a representation learning-based feature extraction for robust features from the reconstructed and latent image spaces. It utilizes a locally invariant segmented-based fractal texture descriptor to extract scale, rotation, and translation-invariant features from the regions of interest (ROIs) in fingerprint and iris images. The feature vectors from two modalities are then fused to form a single feature vector. A quantization mechanism is applied to generate a stable and unique key. The efficacy of the proposed method was evaluated through a number of experiments with multimodal datasets. The uniqueness, dissimilarity, and randomness analyses revealed the usefulness of the proposed approach against several security threats. Indeed, the proposed method offers a promising solution for generating secure keys from multimodal biometric traits.