Time Series Based for Online Signature Verification

Abstract

Signature verification system is to match the tested signature with a claimed signature. This paper propose time series based for feature extraction method and dynamic time warping for match method. The system made by process of testing 900 signatures belong to 50 participants, 3 signatures for reference and 5 signatures from original user, simple imposters and trained imposters for signatures test. The final result system was tested with 50 participants with 3 references. This test obtained that system accuracy without imposters is 90,45 % at threshold 44 with rejection errors (FNMR) is 5,2% and acceptance errors (FMR) is 4,35 %, when with imposters system accuracy is 80,13 % at threshold 27 with error rejection (FNMR) is 15,6% and acceptance errors (average FMR) is 4,26 %, with details as follows: acceptance errors is 0,39%, acceptance errors simple imposters is 3,2% and acceptance errors trained imposters is 9,2%.