Biometric Identification Advances: Unimodal to Multimodal Fusion of Face, Palm, and Iris Features

Abstract

Due to increased information security concerns, biometric recognition technology has become more important. Unimodal biometrics still work effectively, but they struggle with noise sensitivity and spoof attack susceptibility since they rely on a single data source. This paper uses advances in deep learning and machine learning to propose new unimodal systems for the palm, face, and iris. These models use deep wavelet transform networks (WTN) for face and iris identification and deep convolutional neural networks (CNNs) for palmprint identification. In addition, we introduce a novel multimodal biometric system based on unimodal systems. We get 98.29% for face, 98.86% for palmprint, and 95.59% for iris in individual unimodal systems with Support Vector Machines (SVM). This is done by using the new property MULB dataset, which has many biometric features. The multimodal system achieves 99.88% accuracy and a 0.0186 equal error rate, underscoring the relevance of several biometric features and the superior performance of the identification system.

Keywords

• deep learning
• feature extraction
• feature level fusion
• multimodal biometrics identification
• machine learning