System architecture for secure mobile internet voting
This thesis focuses on the development of an enhanced innovative secure mobile Internet voting system architecture that offers desirable security requirements to theoretically mitigate some of the intrinsic administrative and logistical challenges of voting, inter alia lack of mobility support for voters, voter inconvenience, election misconduct, and possible voter coercion often associated with the conventional poll-site voting system. Systems in existence have tended to revolve around the need to provide ubiquitous voting, but lack adequate control mechanism to address, in particular, the important security requirement of controlling possible coercion in ubiquitous voting. The research work reported in this thesis improves upon a well-developed Sensus reference architecture. It does so by leveraging the auto-coupling capability of near field communication, as well as the intrinsic merits of global positioning system, voice biometric authentication, and computational intelligence techniques. The leveraging of the combination of these features provides a theoretical mitigation of some of the security challenges inherent in electoral systems previously alluded to. This leveraging also offers a more pragmatic approach to ensuring high level, secure, mobile Internet voting such as voter authentication. Experiments were performed using spectral features for realising the voice biometric based authentication of the system architecture developed. The spectral features investigated include Mel-frequency Cepstral Coefficients (MFCC), Mel-frequency Discrete Wavelet Coefficients (MFDWC), Linear Predictive Cepstral Coefficients (LPCC), and Spectral Histogram of Oriented Gradients (SHOG). The MFCC, MFDWC and LPCC usually have higher dimensions that oftentimes lead to high computational complexity of the pattern matching algorithms in automatic speaker authentication systems. In this study, higher dimensions of each of the features were reduced per speaker using Histogram of Oriented Gradients (HOG) algorithm, while neural network ensemble was utilised as the pattern-matching algorithm. Out of the four spectral features investigated, the LPCC-HOG gave the best statistical results with an R statistic of 0.9257 and Mean Square Error of 0.0361. These compact LPCC-HOG features are highly promising for implementing the authentication module of the secure mobile Internet voting system architecture reported in this thesis.