(2011) Dynamic Signature Verification and The HBSI Model
DYNAMIC SIGNATUREVERIFICATION AND THEHBSI MODELPurdue University: Michael Brockly | Stephen ElliottUniversity of Kent: Richard Guest | James Scott
RESEARCH QUESTION • How can the Human-Biometric Sensor Interaction (HBSI) Model be further updated for behavioral biometrics? • Can this be done for behavioral biometrics, especially dynamic signature verification?
RESPONSIVE • Currently the model has been created for physiological biometrics and tested most extensively with fingerprint. • As behavioral biometrics become more trusted, the HBSI model needs to stay current.
RELEVANT • Signing as a proof of identity is no new idea • Dynamic Signatures can be a strong anti-theft device
BEHAVIORAL BIOMETRICS • Signature, keystroke, voice, mouse movement • Different from physiological biometrics • Increase in variability
DYNAMIC SIGNATUREVERIFICATION (DSV) • Use digitizer to capture interaction • Common metrics: - min/max/avg velocity - in both x and y dimensions - pen up / pen down time - size of signature - total pen distance - more
2 DIGITIZERS • Virtual Ink - ePad Ink • Paper and Ink - ePad
INCORRECT PRESENTATIONS • Incorrect presentations are errors caused by the signer at point of transaction • These errors include: - Defective Interactions - Concealed Interactions - False Interactions
DEFECTIVE INTERACTIONS Signature is deemed incorrect and no signature channel data is collected • Pen pressure too light at all sample points in paper and ink system • No virtual ink is presented as feedback to the signer in a virtual ink system • Latency or temporal error causes data to not be recorded • User abandons process before data is collected
CONCEALED INTERACTIONS Signature is deemed incorrect and is not detected by the system • User not satisfied but does not abandon • Signature not captured due to latency • Signature out of bounds • Pen pressure or ink dispersal incorrect
FALSE INTERACTIONS Signature is deemed incorrect or unrepresentative by the signer with sample points recorded • Repetitive motion due to faulty ink flow or varying pressure • Repetitive motion due to errors by the signer • Incomplete/unrepresentative signature due to bodily movements • Pen interface issues
CORRECT PRESENTATIONS • Errors can still occur despite a correct presentation: - Failure to Detect - Failure to Process • If no errors occur: - Successfully Processed Sample
FAILURE TO DETECT Occur when the signer has donated a signature that they have deemed acceptable but is not detected by the biometric system • Pen pressure is too light across all sample points even though ink (virtual or physical) is left on the surface/digitizer • No samples are recorded due to latency
FAILURE TO PROCESS Occur when the signature has been deemed acceptable and detected by the system but not considered as an accurate representation • Latency or other temporal errors result in partially captured signature • Start and/or end segments are missing • Part of signature it outside capture area - Occurs if boundary areas are not defined • Pen pressure is too light across all sample points even though ink (virtual or physical) is left on the surface/digitizer
FTP CONTINUED • Pen pressure is variable and too low in some areas although though ink (virtual or physical) is left on the surface/digitizer • Pen malfunction causes channel data disruption - Barrel button pressed during donation • System performance fails to capture certain parts of a signature - Machine with too low of free memory
SUCCESSFULLY PROCESSEDSAMPLE (SPS) This is the result of a correct presentation that both the user and biometric system have deemed correct
RESULTS • 10 potential error points in both the paper and ink system, as well as the virtual ink system • Paper and Ink System - Potential for 1 FTD, 3 FTP, 4 FI, 3 CI, and 1 DI • Virtual Ink System - Potential for 1 FTD, 2 FTP, 6 FI, 6 CI, and 1 DI
FUTURE WORK • Data collection to further test this model • Examination of non-genuine users • Test with more behavioral biometrics - Voice, keystroke, etc.
CONTACT INFORMATION Michael Brockly • Undergraduate Researcher at BSPA Lab • firstname.lastname@example.org Richard Guest PhD • Senior Lecturer at University of Kent • email@example.com Stephen Elliott PhD • Associate Professor at BSPA Lab • firstname.lastname@example.org James Scott • Graduate Researcher at University of Kent • email@example.com