M.Sc. Tezi Görüntüleme | |||||||||||||||||||||
|
|
||||||||||||||||||||
| Summary: Various biometric verification systems are used nowadays to ensure information security. One of them is swipe biometric, which is a recently popular touchscreen biometrics. This biometric, unlike other physiological biometrics, does not require any additional hardware and intends to continuously authenticate the user from the swipe gestures that performed on touch screens in the background. In this study, it is aimed to develop approaches that will reveal the characteristic differences of swipe gestures between the individuals. Firstly, the effect of increasing the number of statistical features obtained from the raw swipe data on the performance was investigated. After that, state of the art feature extraction methods such as KLPP, SR KDA, KPCA, and KDA and commonly used classification methods such as LS SVM, GMM, MLP and WK NN were implemented to evaluate the efficiency of them on the swipe biometric verification systems. Also, the effect of combining classification methods at the score level, using multiple swipe samples instead of a single swipe sample, and pattern based verifying with CNN from the artificial images generated from raw data with different techniques were examined. For the performance analysis, 3 different data sets (Serwadda, JSS18 and Umdaa) with different characteristics which are widely used in the literature were used.
Key Words: Active Authentication, Swipe Biometrics, Feature Extraction. | |||||||||||||||||||||