Experimental Analysis of Biometric System using Various Multimodal Fusion Algorithms

Abstract

Demand for high end privacy and security in human computer interaction, telecom environment is very high in the era of digital world. Multibiometric system combines information from multiple biometric traits of an individual and has an exceptional ability to address these demands with add-on customer satisfaction. It also overcomes intra class variations, non-universality, noisy data and attacks during authentication process. This paper proposes a multibiometric system suitable for secure access of data, devices and services. A database has been constructed using real time multiple biometric samples acquired from 500 individuals in an unconstrained environment. Existence of noise in the samples captured in an unconstrained environment are removed using filtering techniques, and the contrast is adjusted using dark channel priorities and scattering model. Then, the region of interest and features appropriate to each trait are extracted and fused in various forms like multiple samples, instances and traits in recognizing an individual. The proposed system is analysed by computing genuine and false acceptance rates. With the promising experimental results of various fusion schemes, the authentication is tested using transfer learning process with automatic extraction of essential features using Convolution Neural Network and classifying the target using Support Vector Machine (SVM), which outperforms in identifying an individual through fusion of biometric features acquired even in an unconstrained environment. Hence this authentication process could be modified into an effective one to identify and monitor the user interacting with a security related application in online mode with their unique available unconstrained features.