Multimodal biometrics scheme based on discretized eigen feature fusion for identical twins identification

Abstract

The subject of twins multimodal biometrics identification (TMBI) has consistently been an interesting and also a valuable area of study. Considering high dependency and acceptance, TMBI greatly contributes to the domain of twins identification in biometrics traits. The variation of features resulting from the process of multimodal biometrics feature extraction determines the distinctive characteristics possessed by a twin. However, these features are deemed as inessential as they cause the increase in the search space size and also the difficulty in the generalization process. In this regard, the key challenge is to single out features that are deemed most salient with the ability to accurately recognize the twins using multimodal biometrics. In identification of twins, effective designs of methodology and fusion process are important in assuring its success. These processes could be used in the management and integration of vital information including highly selective biometrics characteristic possessed by any of the twins. In the multimodal biometrics twins identification domain, exemplification of the best features from multiple traits of twins and biometrics fusion process remain to be completely resolved. This research attempts to design a new scheme and more effective multimodal biometrics twins identification by introducing the Dis-Eigen feature-based fusion with the capacity in generating a uni-representation and distinctive features of numerous modalities of twins. First, Aspect United Moment Invariant (AUMI) was used as global feature in the extraction of features obtained from the twins handwritingfingerprint shape and style. Then, the feature-based fusion was examined in terms of its generalization. Next, to achieve better classification accuracy, the Dis-Eigen feature-based fusion algorithm was used. A total of eight distinctive classifiers were used in executing four different training and testing of environment settings. Accordingly, the most salient features of Dis-Eigen feature-based fusion were trained and tested to determine the accuracy of the classification, particularly in terms of performance. The results show that the identification of twins improved as the error of similarity for intra-class decreased while at the same time, the error of similarity for inter-class increased. Hence, with the application of diverse classifiers, the identification rate was improved reaching more than 93%. It can be concluded from the experimental outcomes that the proposed method using Receiver Operation Characteristics (ROC) considerably increases the twins handwriting-fingerprint identification process with 90.25% rate of identification when False Acceptance Rate (FAR) is at 0.01%. It is also indicated that 93.15% identification rate is achieved when FAR is at 0.5% and 98.69% when FAR is at 1.00%. The new proposed solution gives a promising alternative to twins identification application.