Using an Accurate Multimodal Biometric for Human Identification System via Deep Learning

Abstract

Biometric systems for automated identification of an individual rely on behavioral or physiological variables linked with the individual. Biometric systems function in two modes: verification and identification. In the verification mode, a claimed identity is either denied or accepted, and in the identification mode, the identity of an unknown person is described. Multibiometric systems are used to establish an individual's identification by combining information supplied by several biometric sensors, samples, units, algorithms, or features. Multibiometric is an interesting and exciting research topic. It is used to identify people to improve security. Therefore, these systems are intended to prevent spoofing, facilitate continuous monitoring, enhance population coverage, and provide fault tolerance to biometric applications. This study proposes an identification system for the individual based on the ear and tongue pattern. Convolution Neural Network (CNN) extracts the essential features from the input images. This system is robust to noise, brightness variations and insensitive to rotation variation. The proposed method consists of four main stages (i.e., pre-processing, fusion, feature extraction, and finally, classification stage). The proposed method was tested on real datasets and achieved an average accuracy of 99.72% for all datasets.