Biometrics for Mobile Devices: A Comparison of Pattern Classification Approaches

Mobile devices have become indispensable tools nowadays. With growing technologies, applications and services are being added to the mobile devices all the time. Its usage in business and enterprises needed it to be secure from unauthorised access of the device and the services. The extent of protection currently available in those devices is not able to secure it. Biometrics has the privilege of providing secure authentication. The success of biometric authentication and reports of theft and loss of information paved the way in its research on mobile devices. Several performance issues are to be considered when implementing biometrics in mobile devices This report focuses on the processing time and storage requirements for biometrics on mobile devices. A detailed study on different algorithms employed by Face, Fingerprint, Keystroke and Signature biometric techniques has been performed. Most of the algorithms that are used for authentication follows similar approach regardless of the techniques and are broadly categorised between statistical and neural network approaches. Processing time in each approach is spent for feature extraction and classification and the storage for holding these features. Neural network techniques performs authentication with higher accuracy but require huge memory capacity and longer training time which makes it infeasible to be employed in mobile devices. Statistical approaches although requires less processing time than neural networks, still requires considerable processing time to perform authentication in real time. Biometric authentication can become possible in mobile devices if the current development in hardware and software technologies such as floating point processors, DSP instruction sets, low power sensors and code optimised software will be implemented fully on these devices.School of Computing, Communications and Electronic