ADAPTIVE FEATURE-LEVEL FUSION FOR TRI-MODAL BIOMETRIC RECOGNITION USING AN ENHANCED CLONAL MATCHING ALGORITHM

Abstract

Over the past decade, biometric recognition systems have emerged as highly effective for identifying individuals and have attracted significant attention for their reliability and wide range of applications in authentication. However, unimodal biometric systems, which rely on a single biometric trait, often face limitations in terms of performance and accuracy. These shortcomings can be addressed by integrating multiple physiological or behavioral traits, thereby creating a multimodal biometric system that enhances both identification and verification. In this study, a tri-modal biometric recognition system was developed that incorporates facial, iris, and signature traits. Feature-level fusion was employed using an Enhanced Clonal Matching Algorithm (ECMA), while classification was performed through a parallel multi-Support Vector Machine (SVM) setup utilizing a quadratic kernel function. The decision-making process relied on thresholding techniques to finalize classifications. Implementation was carried out using multimodal biometric images of 200 participants on MATLAB platform. Performance evaluations of the system using Face, Iris, and Signature data were conducted at a threshold value of 0.75, comparing results from standard CMA and ECMA methods. The obtained result values for true positives, false negatives, false positives, true negatives, false positive ratio (%), accuracy (%), sensitivity (%), specificity (%), precision (%), and processing time (in seconds) were 173, 176; 7, 4; 4, 1; 56, 59; 6.67, 1.67; 95.42, 97.92; 96.11, 97.78; 93.33, 98.33; 97.74, 99.44; and 84.51 sec., 75.81 sec., respectively. The proposed ECMA demonstrated an adaptive matching rate strategy, contributing to its overall effectiveness, which ultimately could enhance secure access control to sensitive areas and valuable assets. 3D biometric images might be considered in the future.