Multimodal Biometrics Using Features from Face, Iris, and Conjunctival Vasculature

Abstract

The performance deficiencies of unimodal biometrics systems arising out of background noise, signal noise and distortions, and environment or device variations can be overcome by using a multimodal biometric system. In this study, an attempt has been made to implement a multimodal biometric fusion using the modalities of face, iris, and conjunctival vasculature. The features from each modality are first extracted using a local ternary pattern-based texture descriptor, viz., Steady Illumination colour Local Ternary Pattern, and then, these are fused at the feature level. The fusion is implemented using a concatenation of the extracted features from these modalities and similarity matching is conducted using Zero-Mean Sum of Squared Differences. Furthermore, the homogeneity of the features is maintained by applying the same feature extractor for all the modalities. The feature extractor is sufficiently efficient in extracting all important features to be fused, which leads to better accuracy. Moreover, the increased number of feature sets because of the fusion at the feature level is reduced using a genetic algorithm, which further improves the accuracy. Experimental results show the effectiveness of the proposed method and results reveal that multimodal biometrics verification is considerably more reliable and precise than the single biometric approach.