Multimodal Biometric Authentication Based on Advanced Data Mining & Machine Learning Techniques

Main Article Content

B. Karthikeyan, D. Suryaprabha, B. Narasimhan, S. Manikandan

Abstract

Biometric is emerging technology in identification and authentication of human being with more reliable and accurate. It is hard to imitate, forge, share, distribute and cannot be stolen, forgotten. Combining multiple biometric systems is a promising solution to provide more security. It eliminates the disadvantages of unimodal biometric systems such as non-universality, noise in sensed data, intra-class variations, distinctiveness, spoof attacks and traditional method of authenticating a human and their identity. The proposed methods in this research depicts a multimodal biometric algorithm which is designed to recognize individuals for robust and secured authentication using normalized score level fusion techniques for optimization in order to reduce False Acceptance Rate and False Rejection Rate and to enhance accuracy. In this research work, the multimodal biometric algorithm integrates Iris and Finger Print biometric traits for their best biometric characteristics. Each biometric trait is adapted for preprocessing techniques such as localization and normalization, before recognition in order to improve the image quality and recognition rate, each trait is recognized by individual recognition algorithm. Matching algorithm provides score and the score is normalized before fusion. Normalization brings the homogeneity for score to apply fusion rule, because in multimodal biometric environment different modalities produce heterogeneous scores. Score level fusion approach is applied to integrate scores from different multimodal biometrics and optimized using Machine Learning Algorithms for robust authentication, enhanced security and accuracy. Here MATLAB is used for implementation. The performance of the algorithm is evaluated by FVC-2004 Dataset for fingerprint and CASIA Dataser for Iris. The database includes multimodal data from 106 individuals. The database is obtained with authenticated agreement from the research website experimental analysis.

Article Details

Section
Articles